| <?xml version="1.0" encoding="iso-8859-1"?> |
| <!DOCTYPE Meeting SYSTEM "http://www.icsi.berkeley.edu/Speech/mr/Meeting1.dtd"> |
|
|
| <Meeting Session="Bro004" DateTimeStamp="2000-11-30-1050" Version="2" VersionDate="Jan 7 2004"> |
|
|
| <Preamble> |
| <Notes> |
| Channel 9 (me006) appears to have a loose connection. For example, |
| while reading the digits, he pops in and out. This causes extremely |
| loud spikes. However, his digits are clearly audible on PZM1 (channel E). |
| For this reason they are transcribed as un-miked when there are trouble |
| spots. |
| |
| me011 is audible prior to digits task for several utterances, but |
| doesn't have a close-talking microphone, and is transcribed un-miked. |
| Not a regular meeting participant. |
| </Notes> |
| <Channels> |
| <Channel Name="chan0" Mic="l1" Transmission="w1" Gain="8" AudioFile="chan0.sph"/> |
| <Channel Name="chan1" Mic="s1" Transmission="w1" Gain="8" AudioFile="chan1.sph"/> |
| <Channel Name="chan2" Mic="s1" Transmission="w1" Gain="8" AudioFile="chan2.sph"/> |
| <Channel Name="chan3" Mic="s1" Transmission="w1" Gain="8" AudioFile="chan3.sph"/> |
| <Channel Name="chan4" Mic="s1" Transmission="w1" Gain="8" AudioFile="chan4.sph"/> |
| <Channel Name="chan6" Mic="c2" Transmission="u1" Gain="8" AudioFile="chan6.sph"/> |
| <Channel Name="chan7" Mic="c2" Transmission="u1" Gain="8" AudioFile="chan7.sph"/> |
| <Channel Name="chan9" Mic="a1" Transmission="j1" Gain="11" AudioFile="chan9.sph"/> |
| <Channel Name="chanB" Mic="p1" Transmission="j1" Gain="12" AudioFile="chanB.sph"/> |
| <Channel Name="chanC" Mic="u3" Transmission="j1" Gain="12" AudioFile="chanC.sph"/> |
| <Channel Name="chanD" Mic="u3" Transmission="j1" Gain="12" AudioFile="chanD.sph"/> |
| <Channel Name="chanE" Mic="c2" Transmission="u1" Gain="8" AudioFile="chanE.sph"/> |
| <Channel Name="chanF" Mic="c2" Transmission="u1" Gain="8" AudioFile="chanF.sph"/> |
| </Channels> |
| <Participants DB="icsi1.spk"> |
| <Participant Name="me018" Channel="chan0"/> |
| <Participant Name="me013" Channel="chan1"/> |
| <Participant Name="me034" Channel="chan2"/> |
| <Participant Name="mn007" Channel="chan3"/> |
| <Participant Name="fn002" Channel="chan4"/> |
| <Participant Name="me006" Channel="chan9"/> |
| <Participant Name="me026" Channel="chanB"/> |
| <Participant Name="me011"/> |
| </Participants> |
| </Preamble> |
|
|
|
|
| <Transcript StartTime="0.0" EndTime="4207.686"> |
|
|
| <Segment StartTime="0.000" EndTime="0.423" Participant="me018"> |
| <VocalSound Description="sings low note "uh""/> <Comment Description="testing mike"/> |
| </Segment> |
| <Segment StartTime="0.388" EndTime="7.752" CloseMic="false"> |
| <Comment Description="testing microphones. Meeting starts after 33 seconds."/> |
| </Segment> |
| <Segment StartTime="0.405" EndTime="2.040" Participant="me013"> |
| O_K. <VocalSound Description="outbreath"/> |
| </Segment> |
| <Segment StartTime="0.567" EndTime="1.822" Participant="me034"> |
| Oh, I don't - |
| </Segment> |
| <Segment StartTime="2.470" EndTime="3.560" Participant="me018"> |
| I think I'm zero. |
| </Segment> |
| <Segment StartTime="2.834" EndTime="4.107" Participant="me013"> |
| Wow! <VocalSound Description="laugh"/> |
| </Segment> |
| <Segment StartTime="4.130" EndTime="4.850" Participant="fn002"> |
| Ah- |
| </Segment> |
| <Segment StartTime="4.211" EndTime="6.885" Participant="me034"> |
| Hello, hello, hello, hello. |
| </Segment> |
| <Segment StartTime="4.251" EndTime="6.702" Participant="me006"> |
| Wh- what causes the crash? |
| </Segment> |
| <Segment StartTime="4.292" EndTime="5.616" Participant="me013"> |
| <Emphasis> Unprecedented. </Emphasis> |
| </Segment> |
| <Segment StartTime="5.501" EndTime="6.712" Participant="me018"> |
| Did you fix something? |
| </Segment> |
| <Segment StartTime="7.128" EndTime="7.963" Participant="me034"> |
| Hello. |
| </Segment> |
| <Segment StartTime="7.156" EndTime="10.444" Participant="me013"> |
| <VocalSound Description="laugh"/> |
| </Segment> |
| <Segment StartTime="7.752" EndTime="8.786" CloseMic="false"> |
| <NonVocalSound Description="door slams"/> |
| </Segment> |
| <Segment StartTime="9.182" EndTime="10.450" Participant="fn002"> |
| Five, five. |
| </Segment> |
| <Segment StartTime="9.433" EndTime="14.106" Participant="me006"> |
| Oh, maybe it's the turning - turning off and turning on of the mike, right? |
| </Segment> |
| <Segment StartTime="9.461" EndTime="12.050" Participant="me034"> |
| Hello, hello. |
| </Segment> |
| <Segment StartTime="17.777" EndTime="18.329" CloseMic="false"> |
| <NonVocalSound Description="door slams"/> |
| </Segment> |
| <Segment StartTime="20.260" EndTime="21.328" Participant="me006"> |
| <VocalSound Description="clear throat"/> |
| </Segment> |
| <Segment StartTime="21.854" EndTime="22.640" Participant="me006"> |
| <VocalSound Description="cough"/> |
| </Segment> |
| <Segment StartTime="22.336" EndTime="24.175" Participant="me013"> |
| Uh, you think that's <Emphasis> you? </Emphasis> |
| </Segment> |
| <Segment StartTime="23.514" EndTime="27.001" Participant="me006"> |
| <VocalSound Description="loud cough"/> Yeah, O_K, mine's working. |
| </Segment> |
| <Segment StartTime="23.914" EndTime="30.514" Participant="me034"> |
| Aaa-aaa-aaa. <Comment Description="nonword, singing into mike to find channel"/> O_K. That's me. |
| </Segment> |
| <Segment StartTime="24.175" EndTime="24.560" Participant="me013"> |
| <Uncertain> Oh. </Uncertain> |
| </Segment> |
| <Segment StartTime="34.314" EndTime="35.378" Participant="me013"> |
| O_K. |
| </Segment> |
| <Segment StartTime="36.662" EndTime="39.943" Participant="me013"> |
| O_K. So, um I guess we are |
| </Segment> |
| <Segment StartTime="39.943" EndTime="42.906" Participant="me013"> |
| <Pause/> um <Pause/> gonna do the digits at the end. |
| </Segment> |
| <Segment StartTime="42.906" EndTime="44.836" Participant="me013"> |
| <Pause/> Uh <Pause/> |
| </Segment> |
| <Segment StartTime="44.737" EndTime="48.010" Participant="mn007"> |
| Channel - channel three, yeah. O_K. |
| </Segment> |
| <Segment StartTime="47.317" EndTime="48.994" Participant="fn002"> |
| Mmm, channel five? |
| </Segment> |
| <Segment StartTime="47.599" EndTime="48.748" Participant="me034"> |
| Channel two. |
| </Segment> |
| <Segment StartTime="48.994" EndTime="50.111" Participant="fn002"> |
| Doesn't work? |
| </Segment> |
| <Segment StartTime="49.760" EndTime="50.900" Participant="me034"> |
| Two. |
| </Segment> |
| <Segment StartTime="50.018" EndTime="52.091" Participant="me013"> |
| Yeah, <Emphasis> that's </Emphasis> the <Emphasis> mike </Emphasis> number <Emphasis> there, </Emphasis> |
| </Segment> |
| <Segment StartTime="51.657" EndTime="52.886" Participant="me006"> |
| <VocalSound Description="clear throat"/> |
| </Segment> |
| <Segment StartTime="52.091" EndTime="52.880" Participant="me013"> |
| uh <Pause/> |
| </Segment> |
| <Segment StartTime="53.007" EndTime="53.731" Participant="me006"> |
| <VocalSound Description="clear throat"/> |
| </Segment> |
| <Segment StartTime="53.552" EndTime="54.223" Participant="fn002"> |
| No? |
| </Segment> |
| <Segment StartTime="53.964" EndTime="55.591" Participant="me018"> |
| Is it written on her <Emphasis> sheet, </Emphasis> I believe. |
| </Segment> |
| <Segment StartTime="54.204" EndTime="57.445" Participant="mn007"> |
| Mike four. <VocalSound Description="blows into mike, whistles"/> |
| </Segment> |
| <Segment StartTime="54.546" EndTime="59.961" Participant="me006"> |
| <VocalSound Description="blows into mike"/> Watch this. <VocalSound Description="loud cough"/> <Comment Description="testing mike"/> Yep, that's me. |
| </Segment> |
| <Segment StartTime="57.059" EndTime="58.861" Participant="me013"> |
| Uh, mike number five, |
| </Segment> |
| <Segment StartTime="57.700" EndTime="60.278" Participant="fn002"> |
| Ah, <Foreign Language="es" Description="literally "It was the four", meaning "It's channel four""> era el cuatro. </Foreign> Yeah. |
| </Segment> |
| <Segment StartTime="59.674" EndTime="60.324" Participant="me013"> |
| and <Pause/> |
| </Segment> |
| <Segment StartTime="59.974" EndTime="60.968" Participant="me018"> |
| But, <Emphasis> channel </Emphasis> <Pause/> <Comment Description="providing lead-in"/> |
| </Segment> |
| <Segment StartTime="60.324" EndTime="61.712" Participant="me013"> |
| channel - channel four. |
| </Segment> |
| <Segment StartTime="60.880" EndTime="62.069" Participant="fn002"> |
| Yeah yeah yeah. |
| </Segment> |
| <Segment StartTime="61.712" EndTime="62.677" Participant="me013"> |
| <Pause/> This is <Emphasis> you. </Emphasis> |
| </Segment> |
| <Segment StartTime="62.795" EndTime="64.034" Participant="fn002"> |
| O_K. I saw that. <Comment Description="whispered"/> |
| </Segment> |
| <Segment StartTime="65.956" EndTime="68.067" Participant="fn002"> |
| Ah - yeah, it's O_K. |
| </Segment> |
| <Segment StartTime="66.321" EndTime="66.912" Participant="me013"> |
| Yeah. |
| </Segment> |
| <Segment StartTime="66.912" EndTime="70.797" Participant="me013"> |
| <Pause/> And I'm channel uh <Emphasis> two </Emphasis> I think, <Pause/> or channel - |
| </Segment> |
| <Segment StartTime="67.823" EndTime="69.272" Participant="me034"> |
| Ooo. <Comment Description="nonword, singing into mike to find channel"/> |
| </Segment> |
| <Segment StartTime="70.374" EndTime="71.713" Participant="me034"> |
| I think <Emphasis> I'm </Emphasis> channel two. |
| </Segment> |
| <Segment StartTime="71.509" EndTime="74.378" Participant="me013"> |
| Oh, I'm channel - must be channel one. Channel one? <Comment Description="loudly, looking at display"/> Yes, O_K. |
| </Segment> |
| <Segment StartTime="72.784" EndTime="73.563" Participant="fn002"> |
| Channel - |
| </Segment> |
| <Segment StartTime="75.439" EndTime="76.483" Participant="fn002"> |
| <VocalSound Description="clear throat"/> |
| </Segment> |
| <Segment StartTime="76.254" EndTime="78.516" Participant="me013"> |
| O_K. So uh |
| </Segment> |
| <Segment StartTime="78.516" EndTime="84.316" Participant="me013"> |
| <Pause/> I also copied uh the results that we all got in the mail I think from uh - |
| </Segment> |
| <Segment StartTime="84.316" EndTime="87.894" Participant="me013"> |
| <Pause/> from O_G_I and we'll go - go through them also. |
| </Segment> |
| <Segment StartTime="87.894" EndTime="90.995" Participant="me013"> |
| <Pause/> So where are we on - |
| </Segment> |
| <Segment StartTime="90.995" EndTime="93.326" Participant="me013"> |
| <Pause/> on uh <VocalSound Description="laugh"/> |
| </Segment> |
| <Segment StartTime="93.326" EndTime="95.599" Participant="me013"> |
| <Pause/> our runs? |
| </Segment> |
| <Segment StartTime="95.581" EndTime="105.042" Participant="mn007"> |
| Uh so. <Pause/> uh - We - So <Pause/> As I was already said, we - we mainly focused on |
| </Segment> |
| <Segment StartTime="107.511" EndTime="119.333" Participant="mn007"> |
| uh four kind of features. The P_L_P, the P_L_P with J_RASTA, the M_S_G, and the M_F_C_C from the baseline Aurora. |
| </Segment> |
| <Segment StartTime="108.418" EndTime="109.843" Participant="me013"> |
| Excuse me. <Comment Description="whispered"/> |
| </Segment> |
| <Segment StartTime="119.700" EndTime="121.794" Participant="fn002"> |
| I decided to talk about that. <Comment Description="whispered to [mn007]"/> |
| </Segment> |
| <Segment StartTime="120.884" EndTime="121.549" Participant="me013"> |
| Mm-hmm. |
| </Segment> |
| <Segment StartTime="122.002" EndTime="126.982" Participant="mn007"> |
| Uh, and we focused for the - the test part on the English and the Italian. |
| </Segment> |
| <Segment StartTime="129.518" EndTime="148.537" Participant="mn007"> |
| Um. We've trained uh several neural networks on - so - on the T_I-digits English <Pause/> and on the Italian data and also on the broad uh <Pause/> English uh French and uh Spanish databases. |
| </Segment> |
| <Segment StartTime="149.875" EndTime="168.215" Participant="mn007"> |
| Mmm, so there's our result tables here, for the tandem approach, and um, actually what we - we <Uncertain NumSyllables="1"> @@ </Uncertain> observed is that if the network is trained on the task data it works pretty well. |
| </Segment> |
| <Segment StartTime="169.855" EndTime="172.772" Participant="me006"> |
| <VocalSound Description="laugh"/> |
| </Segment> |
| <Segment StartTime="169.895" EndTime="171.068" Participant="me034"> |
| <VocalSound Description="breath-laugh"/> |
| </Segment> |
| <Segment StartTime="171.202" EndTime="172.312" Participant="fn002"> |
| <VocalSound Description="laugh"/> |
| </Segment> |
| <Segment StartTime="171.845" EndTime="172.324" Participant="me013"> |
| <VocalSound Description="laugh"/> |
| </Segment> |
| <Segment StartTime="172.243" EndTime="173.573" Participant="me011" CloseMic="false"> |
| I can't get <Emphasis> back </Emphasis> far enough. |
| </Segment> |
| <Segment StartTime="173.017" EndTime="175.443" Participant="fn002"> |
| <VocalSound Description="laugh"/> |
| </Segment> |
| <Segment StartTime="173.485" EndTime="174.716" Participant="me034"> |
| <Uncertain> Chicken on the grill. </Uncertain> |
| </Segment> |
| <Segment StartTime="173.787" EndTime="175.531" Participant="me013"> |
| O_K. Our - our uh - |
| </Segment> |
| <Segment StartTime="175.531" EndTime="178.079" Participant="me013"> |
| <Pause/> There's a - <Pause/> We're pausing for a photo - |
| </Segment> |
| <Segment StartTime="175.716" EndTime="177.203" Participant="me011" CloseMic="false"> |
| Sorry, guys. <VocalSound Description="laugh"/> |
| </Segment> |
| <Segment StartTime="175.913" EndTime="178.082" Participant="fn002"> |
| <VocalSound Description="laugh"/> |
| </Segment> |
| <Segment StartTime="176.601" EndTime="177.736" Participant="me034"> |
| Try <Emphasis> that </Emphasis> corner. |
| </Segment> |
| <Segment StartTime="177.310" EndTime="178.925" Participant="me018"> |
| How about over th- from the front of the room? |
| </Segment> |
| <Segment StartTime="178.925" EndTime="180.164" Participant="me034"> |
| Yeah, it's longer. |
| </Segment> |
| <Segment StartTime="180.164" EndTime="183.081" Participant="me013"> |
| We're pausing for a photo opportunity here. Uh. |
| </Segment> |
| <Segment StartTime="183.081" EndTime="183.605" Participant="me013"> |
| <VocalSound Description="inbreath"/> |
| </Segment> |
| <Segment StartTime="183.605" EndTime="184.165" Participant="me013"> |
| Uh. |
| </Segment> |
| <Segment StartTime="183.716" EndTime="185.659" Participant="me006"> |
| <VocalSound Description="laugh"/> |
| </Segment> |
| <Segment StartTime="184.533" EndTime="186.036" Participant="fn002"> |
| <VocalSound Description="laugh"/> |
| </Segment> |
| <Segment StartTime="186.261" EndTime="186.658" Participant="me013"> |
| So. |
| </Segment> |
| <Segment StartTime="186.708" EndTime="190.979" Participant="me006"> |
| Oh wait wait wait wait wait. Wait. Hold on. Hold on. Let me give you a black screen. |
| </Segment> |
| <Segment StartTime="187.362" EndTime="188.576" Participant="me034"> |
| Get out of the - Yeah. |
| </Segment> |
| <Segment StartTime="188.700" EndTime="189.509" Participant="me013"> |
| O_K. |
| </Segment> |
| <Segment StartTime="191.432" EndTime="193.497" Participant="me034"> |
| <VocalSound Description="laugh"/> |
| </Segment> |
| <Segment StartTime="194.555" EndTime="197.435" Participant="me013"> |
| <VocalSound Description="laugh"/> |
| </Segment> |
| <Segment StartTime="195.359" EndTime="196.157" Participant="me011" CloseMic="false"> |
| One more. |
| </Segment> |
| <Segment StartTime="196.480" EndTime="199.017" Participant="fn002"> |
| <VocalSound Description="laugh"/> |
| </Segment> |
| <Segment StartTime="197.435" EndTime="198.404" Participant="me013"> |
| He's facing <Emphasis> this </Emphasis> way. |
| </Segment> |
| <Segment StartTime="198.047" EndTime="200.586" Participant="me011" CloseMic="false"> |
| Because we <Emphasis> said </Emphasis> we were gonna <Emphasis> do </Emphasis> this and I <Emphasis> just </Emphasis> remembered. |
| </Segment> |
| <Segment StartTime="198.211" EndTime="200.276" Participant="me006"> |
| <VocalSound Description="laugh"/> |
| </Segment> |
| <Segment StartTime="198.404" EndTime="199.960" Participant="me013"> |
| <VocalSound Description="laugh"/> |
| </Segment> |
| <Segment StartTime="199.522" EndTime="201.373" Participant="fn002"> |
| <VocalSound Description="laugh"/> |
| </Segment> |
| <Segment StartTime="199.960" EndTime="201.078" Participant="me013"> |
| <Uncertain> What? </Uncertain> <VocalSound Description="laugh"/> |
| </Segment> |
| <Segment StartTime="205.937" EndTime="210.640" Participant="me013"> |
| O_K, this - this would be a <Pause/> good section for our silence detection. |
| </Segment> |
| <Segment StartTime="206.769" EndTime="209.441" Participant="me006"> |
| O_K. <VocalSound Description="clear throat"/> |
| </Segment> |
| <Segment StartTime="210.640" EndTime="212.085" Participant="me013"> |
| <Pause/> Um |
| </Segment> |
| <Segment StartTime="210.722" EndTime="211.430" Participant="me034"> |
| Mm-hmm. |
| </Segment> |
| <Segment StartTime="210.905" EndTime="212.126" Participant="me018"> |
| <VocalSound Description="laugh"/> |
| </Segment> |
| <Segment StartTime="210.999" EndTime="212.697" Participant="me011" CloseMic="false"> |
| <VocalSound Description="laugh"/> |
| </Segment> |
| <Segment StartTime="212.393" EndTime="213.299" Participant="fn002"> |
| <VocalSound Description="clear throat"/> |
| </Segment> |
| <Segment StartTime="212.473" EndTime="216.693" Participant="me006"> |
| Musical chairs everybody! <VocalSound Description="laugh"/> |
| </Segment> |
| <Segment StartTime="213.784" EndTime="215.006" Participant="me011" CloseMic="false"> |
| <VocalSound Description="laugh"/> |
| </Segment> |
| <Segment StartTime="214.362" EndTime="214.706" Participant="me013"> |
| Oh. |
| </Segment> |
| <Segment StartTime="215.108" EndTime="215.866" Participant="me018"> |
| <VocalSound Description="laugh"/> |
| </Segment> |
| <Segment StartTime="215.664" EndTime="216.097" Participant="me013"> |
| O_K. |
| </Segment> |
| <Segment StartTime="216.097" EndTime="217.733" Participant="me013"> |
| So um, |
| </Segment> |
| <Segment StartTime="217.492" EndTime="218.526" CloseMic="false"> |
| <NonVocalSound Description="door slams"/> |
| </Segment> |
| <Segment StartTime="217.733" EndTime="218.889" Participant="me013"> |
| <Pause/> you were saying |
| </Segment> |
| <Segment StartTime="218.889" EndTime="221.144" Participant="me013"> |
| <Pause/> about the training data - <Pause/> <Comment Description="providing lead-in"/> |
| </Segment> |
| <Segment StartTime="221.149" EndTime="234.240" Participant="mn007"> |
| Yeah, so if the network is trained on the task data um <Pause/> tandem works pretty well. And uh actually we have uh, results are similar |
| </Segment> |
| <Segment StartTime="221.202" EndTime="222.190" Participant="me034"> |
| <VocalSound Description="sips coffee"/> |
| </Segment> |
| <Segment StartTime="221.453" EndTime="221.972" Participant="me013"> |
| Yeah. |
| </Segment> |
| <Segment StartTime="234.240" EndTime="236.812" Participant="me018"> |
| Do you mean if it's trained <Emphasis> only </Emphasis> on - |
| </Segment> |
| <Segment StartTime="236.504" EndTime="238.082" Participant="mn007"> |
| Only on, yeah. <Comment Description="answering question"/> |
| </Segment> |
| <Segment StartTime="237.412" EndTime="240.081" Participant="me018"> |
| On data from just that task, that language? |
| </Segment> |
| <Segment StartTime="239.524" EndTime="240.407" Participant="mn007"> |
| Just that task. |
| </Segment> |
| <Segment StartTime="240.407" EndTime="246.058" Participant="mn007"> |
| But actually we didn't train network on <Pause/> uh both types of data I mean |
| </Segment> |
| <Segment StartTime="246.058" EndTime="252.979" Participant="mn007"> |
| <Pause/> uh <Pause/> phonetically ba- phonetically balanced uh data and task data. |
| </Segment> |
| <Segment StartTime="252.979" EndTime="259.733" Participant="mn007"> |
| <Pause/> We only did <Emphasis> either </Emphasis> task - task data or <Pause/> uh broad <Pause/> data. |
| </Segment> |
| <Segment StartTime="253.037" EndTime="253.823" Participant="me018"> |
| Mmm. |
| </Segment> |
| <Segment StartTime="259.733" EndTime="260.402" Participant="me018"> |
| Mm-hmm. |
| </Segment> |
| <Segment StartTime="261.291" EndTime="262.498" Participant="mn007"> |
| Um <Pause/> |
| </Segment> |
| <Segment StartTime="262.498" EndTime="263.275" Participant="mn007"> |
| Yeah. So, |
| </Segment> |
| <Segment StartTime="263.305" EndTime="263.847" Participant="me013"> |
| <VocalSound Description="inbreath"/> |
| </Segment> |
| <Segment StartTime="263.847" EndTime="264.763" Participant="me013"> |
| So how - I mean - |
| </Segment> |
| <Segment StartTime="264.740" EndTime="265.806" Participant="me018"> |
| So what's th- |
| </Segment> |
| <Segment StartTime="264.763" EndTime="268.359" Participant="me013"> |
| clearly it's gonna be <Emphasis> good </Emphasis> then but the question is how much <Pause/> worse is it |
| </Segment> |
| <Segment StartTime="268.359" EndTime="270.130" Participant="me013"> |
| <Pause/> if you have <Emphasis> broad </Emphasis> data? |
| </Segment> |
| <Segment StartTime="270.130" EndTime="270.455" Participant="me013"> |
| I mean, |
| </Segment> |
| <Segment StartTime="270.455" EndTime="271.503" Participant="me013"> |
| <Pause/> my assump- |
| </Segment> |
| <Segment StartTime="271.503" EndTime="274.738" Participant="me013"> |
| From what I saw from the earlier results, uh I guess last week, |
| </Segment> |
| <Segment StartTime="274.738" EndTime="277.279" Participant="me013"> |
| <Pause/> was that um, |
| </Segment> |
| <Segment StartTime="277.279" EndTime="281.742" Participant="me013"> |
| <Pause/> if you <Pause/> <Emphasis> trained </Emphasis> on one language and <Emphasis> tested </Emphasis> on another, say, that |
| </Segment> |
| <Segment StartTime="281.742" EndTime="284.217" Participant="me013"> |
| <Pause/> the results were - were relatively <Emphasis> poor. </Emphasis> |
| </Segment> |
| <Segment StartTime="284.503" EndTime="286.342" Participant="mn007"> |
| Mmm. Yeah. |
| </Segment> |
| <Segment StartTime="285.384" EndTime="289.096" Participant="me013"> |
| But - but the question is if you <Emphasis> train </Emphasis> on one language |
| </Segment> |
| <Segment StartTime="289.096" EndTime="291.047" Participant="me013"> |
| <Pause/> but you have a <Emphasis> broad </Emphasis> coverage |
| </Segment> |
| <Segment StartTime="291.047" EndTime="292.673" Participant="me013"> |
| <Pause/> and then test in another, |
| </Segment> |
| <Segment StartTime="292.673" EndTime="294.770" Participant="me013"> |
| <Pause/> does that - <Pause/> is that <Emphasis> improve </Emphasis> things |
| </Segment> |
| <Segment StartTime="294.770" EndTime="296.631" Participant="me013"> |
| <Pause/> i- c- in <Emphasis> comparison? </Emphasis> |
| </Segment> |
| <Segment StartTime="297.532" EndTime="299.131" Participant="mn007"> |
| If we use the same language? |
| </Segment> |
| <Segment StartTime="299.060" EndTime="299.915" Participant="me013"> |
| No, no, no. |
| </Segment> |
| <Segment StartTime="299.915" EndTime="300.511" Participant="me013"> |
| <Pause/> Different lang- <Comment Description="answering question"/> |
| </Segment> |
| <Segment StartTime="300.511" EndTime="301.089" Participant="me013"> |
| <Pause/> So |
| </Segment> |
| <Segment StartTime="301.089" EndTime="301.794" Participant="me013"> |
| <Pause/> um |
| </Segment> |
| <Segment StartTime="301.794" EndTime="306.528" Participant="me013"> |
| <Pause/> If you train on T_I-digits <Pause/> and test on <Emphasis> Italian </Emphasis> digits, <Pause/> you do poorly, |
| </Segment> |
| <Segment StartTime="305.589" EndTime="306.196" Participant="mn007"> |
| Mm-hmm. |
| </Segment> |
| <Segment StartTime="306.528" EndTime="307.341" Participant="me013"> |
| <Pause/> let's say. |
| </Segment> |
| <Segment StartTime="308.462" EndTime="310.630" Participant="me013"> |
| I don't have the <Emphasis> numbers </Emphasis> in front of me, so I'm just <Emphasis> imagining. </Emphasis> |
| </Segment> |
| <Segment StartTime="308.706" EndTime="309.234" Participant="mn007"> |
| But - |
| </Segment> |
| <Segment StartTime="310.630" EndTime="313.915" Participant="mn007"> |
| Yeah but I did not uh do that. We - |
| </Segment> |
| <Segment StartTime="311.017" EndTime="311.486" Participant="me013"> |
| E- |
| </Segment> |
| <Segment StartTime="312.965" EndTime="317.284" Participant="me013"> |
| So, you didn't train on <Pause/> TIMIT and test on - <Pause/> on Italian digits, say? |
| </Segment> |
| <Segment StartTime="317.284" EndTime="320.591" Participant="mn007"> |
| No, we did four - four kind of - of testing, actually. |
| </Segment> |
| <Segment StartTime="321.541" EndTime="324.472" Participant="mn007"> |
| The first testing is <Pause/> with task data - |
| </Segment> |
| <Segment StartTime="324.472" EndTime="327.763" Participant="mn007"> |
| <Pause/> So, with nets trained on task data. |
| </Segment> |
| <Segment StartTime="329.564" EndTime="333.461" Participant="mn007"> |
| So for Italian on the Italian speech <Uncertain NumSyllables="2"> @@ </Uncertain>. |
| </Segment> |
| <Segment StartTime="334.380" EndTime="338.228" Participant="mn007"> |
| The second test is trained on a single language |
| </Segment> |
| <Segment StartTime="339.127" EndTime="344.230" Participant="mn007"> |
| um with broad <Emphasis> database, </Emphasis> but the same language as the t- task data. |
| </Segment> |
| <Segment StartTime="344.230" EndTime="344.618" Participant="me013"> |
| O_K. |
| </Segment> |
| <Segment StartTime="344.705" EndTime="349.624" Participant="mn007"> |
| But for Italian we choose Spanish which <Pause/> we assume is close to Italian. |
| </Segment> |
| <Segment StartTime="349.624" EndTime="352.885" Participant="mn007"> |
| The third test is by using, um |
| </Segment> |
| <Segment StartTime="353.633" EndTime="355.666" Participant="mn007"> |
| the three language database |
| </Segment> |
| <Segment StartTime="356.334" EndTime="358.157" Participant="mn007"> |
| and the fourth is |
| </Segment> |
| <Segment StartTime="356.803" EndTime="359.662" Participant="me013"> |
| W- which in - It has three languages. That's including the w- |
| </Segment> |
| <Segment StartTime="359.662" EndTime="360.095" Participant="me013"> |
| the - |
| </Segment> |
| <Segment StartTime="360.095" EndTime="360.710" Participant="me013"> |
| <Pause/> the - |
| </Segment> |
| <Segment StartTime="360.564" EndTime="361.566" Participant="mn007"> |
| This includes - |
| </Segment> |
| <Segment StartTime="361.704" EndTime="363.001" Participant="me013"> |
| the one that it's - <Comment Description="completing [mn007]'s utterance"/> |
| </Segment> |
| <Segment StartTime="362.560" EndTime="363.070" Participant="mn007"> |
| Yeah. <Comment Description="answering question"/> |
| </Segment> |
| <Segment StartTime="363.068" EndTime="363.358" Participant="me018"> |
| <VocalSound Description="inbreath"/> |
| </Segment> |
| <Segment StartTime="363.283" EndTime="363.879" Participant="mn007"> |
| But <Pause/> not |
| </Segment> |
| <Segment StartTime="363.358" EndTime="363.615" Participant="me018"> |
| In- |
| </Segment> |
| <Segment StartTime="364.331" EndTime="365.610" Participant="mn007"> |
| <Emphasis> digits. </Emphasis> I mean it's - |
| </Segment> |
| <Segment StartTime="365.146" EndTime="365.610" Participant="me013"> |
| Right. <Comment Description="affirming [mn007]'s interrupted utterance"/> |
| </Segment> |
| <Segment StartTime="365.419" EndTime="366.898" Participant="me018"> |
| The three languages |
| </Segment> |
| <Segment StartTime="366.898" EndTime="368.831" Participant="me018"> |
| <Pause/> is <Emphasis> not </Emphasis> digits, it's the broad |
| </Segment> |
| <Segment StartTime="368.831" EndTime="370.013" Participant="me018"> |
| <Pause/> data. O_K. |
| </Segment> |
| <Segment StartTime="368.926" EndTime="369.742" Participant="mn007"> |
| Yeah |
| </Segment> |
| <Segment StartTime="369.742" EndTime="378.406" Participant="mn007"> |
| And the fourth test is uh <Pause/> excluding from these three languages the language <Pause/> that is <Pause/> the task language. |
| </Segment> |
| <Segment StartTime="378.406" EndTime="380.105" Participant="me013"> |
| Oh, O_K, yeah, so, that <Emphasis> is </Emphasis> what I wanted to know. |
| </Segment> |
| <Segment StartTime="379.047" EndTime="379.448" Participant="mn007"> |
| Yeah. |
| </Segment> |
| <Segment StartTime="380.105" EndTime="381.692" Participant="me013"> |
| I just wasn't <Emphasis> saying </Emphasis> it very well, I guess. |
| </Segment> |
| <Segment StartTime="381.304" EndTime="382.533" Participant="mn007"> |
| Uh, yeah. |
| </Segment> |
| <Segment StartTime="382.533" EndTime="399.813" Participant="mn007"> |
| So um <Pause/> for uh T_I-digits for ins- example <Pause/> uh when we go from T_I-digits training to <Pause/> TIMIT training <Pause/> uh we lose <Pause/> uh around ten percent, uh. |
| </Segment> |
| <Segment StartTime="400.700" EndTime="401.863" Participant="me013"> |
| Relative. <Comment Description="adding to [mn007]'s point"/> |
| </Segment> |
| <Segment StartTime="400.700" EndTime="403.655" Participant="mn007"> |
| The error rate increase u- of - of - |
| </Segment> |
| <Segment StartTime="402.697" EndTime="402.968" Participant="me013"> |
| Right. |
| </Segment> |
| <Segment StartTime="403.655" EndTime="405.538" Participant="mn007"> |
| of ten percent, relative. |
| </Segment> |
| <Segment StartTime="406.508" EndTime="414.843" Participant="mn007"> |
| So this is not so bad. And then when we jump to the multilingual data it's uh it become worse and, well |
| </Segment> |
| <Segment StartTime="414.184" EndTime="415.467" Participant="me013"> |
| Ab- about how much? |
| </Segment> |
| <Segment StartTime="415.891" EndTime="422.441" Participant="mn007"> |
| Around uh, let's say, <Pause/> twenty perc- twenty percent further. <Pause/> So. <Pause/> Yeah. |
| </Segment> |
| <Segment StartTime="424.895" EndTime="426.084" Participant="me013"> |
| Twenty percent <Emphasis> further? </Emphasis> |
| </Segment> |
| <Segment StartTime="425.670" EndTime="427.949" Participant="mn007"> |
| Twenty to - to thirty percent further. Yeah. |
| </Segment> |
| <Segment StartTime="428.553" EndTime="432.080" Participant="me018"> |
| And so, remind me, the <Emphasis> multilingual </Emphasis> stuff is <Emphasis> just </Emphasis> the broad data. Right? |
| </Segment> |
| <Segment StartTime="432.080" EndTime="432.335" Participant="mn007"> |
| Yeah. |
| </Segment> |
| <Segment StartTime="432.335" EndTime="436.537" Participant="me018"> |
| It's not the digits. So it's the combination of <Pause/> two things there. |
| </Segment> |
| <Segment StartTime="436.537" EndTime="443.936" Participant="me018"> |
| It's <Pause/> removing the <Pause/> task specific <Pause/> training and <Pause/> it's adding other languages. |
| </Segment> |
| <Segment StartTime="441.796" EndTime="442.280" Participant="mn007"> |
| Yeah. |
| </Segment> |
| <Segment StartTime="443.614" EndTime="444.029" Participant="mn007"> |
| Yeah. |
| </Segment> |
| <Segment StartTime="443.936" EndTime="444.634" Participant="me018"> |
| O_K. |
| </Segment> |
| <Segment StartTime="444.029" EndTime="447.968" Participant="mn007"> |
| <Pause/> But the first step is al- already removing the task s- specific from - from - |
| </Segment> |
| <Segment StartTime="445.900" EndTime="447.609" Participant="me018"> |
| Already, right right right. |
| </Segment> |
| <Segment StartTime="447.968" EndTime="449.721" Participant="mn007"> |
| <Pause/> So. And we lose - |
| </Segment> |
| <Segment StartTime="448.563" EndTime="450.721" Participant="me018"> |
| So they were sort of <Emphasis> building </Emphasis> <Pause/> here? O_K? |
| </Segment> |
| <Segment StartTime="449.721" EndTime="450.439" Participant="mn007"> |
| <Pause/> Yeah. <Comment Description="answering question"/> |
| </Segment> |
| <Segment StartTime="451.126" EndTime="457.089" Participant="mn007"> |
| Uh <Pause/> So, basically when it's <Emphasis> trained </Emphasis> on the - the multilingual broad data |
| </Segment> |
| <Segment StartTime="457.089" EndTime="470.882" Participant="mn007"> |
| <Pause/> um or number - so, the - the <Pause/> ratio of our error rates uh with the <Pause/> baseline error rate is around <Pause/> uh one point one. |
| </Segment> |
| <Segment StartTime="470.882" EndTime="471.610" Participant="mn007"> |
| So. |
| </Segment> |
| <Segment StartTime="470.966" EndTime="471.610" Participant="me013"> |
| Yes. |
| </Segment> |
| <Segment StartTime="471.610" EndTime="475.201" Participant="me013"> |
| <VocalSound Description="inbreath"/> And it's something like one point <Emphasis> three </Emphasis> of - of the |
| </Segment> |
| <Segment StartTime="475.201" EndTime="475.921" Participant="me013"> |
| <Pause/> uh - |
| </Segment> |
| <Segment StartTime="475.921" EndTime="478.838" Participant="me013"> |
| I- i- if you compare everything to the first case at the baseline, |
| </Segment> |
| <Segment StartTime="478.838" EndTime="481.022" Participant="me013"> |
| <Pause/> you get something like one point one |
| </Segment> |
| <Segment StartTime="481.022" EndTime="485.467" Participant="me013"> |
| for the - for the using the same language but a different task, and something like one point three |
| </Segment> |
| <Segment StartTime="485.467" EndTime="488.015" Participant="me013"> |
| <Pause/> for three - three languages |
| </Segment> |
| <Segment StartTime="488.006" EndTime="488.650" Participant="mn007"> |
| No no no. |
| </Segment> |
| <Segment StartTime="488.015" EndTime="489.497" Participant="me013"> |
| <Pause/> <Emphasis> broad </Emphasis> stuff. |
| </Segment> |
| <Segment StartTime="489.497" EndTime="495.529" Participant="mn007"> |
| Uh same language we are at uh - for at English at O_ point eight. |
| </Segment> |
| <Segment StartTime="495.529" EndTime="497.236" Participant="mn007"> |
| <Pause/> So it improves, |
| </Segment> |
| <Segment StartTime="497.236" EndTime="498.411" Participant="mn007"> |
| <Pause/> compared to the baseline. |
| </Segment> |
| <Segment StartTime="499.423" EndTime="500.218" Participant="mn007"> |
| But - |
| </Segment> |
| <Segment StartTime="500.218" EndTime="501.284" Participant="mn007"> |
| So. Le- <Emphasis> let </Emphasis> me. |
| </Segment> |
| <Segment StartTime="501.066" EndTime="504.532" Participant="me013"> |
| I - I - I'm sorry. I - I - I meant something different by baseline |
| </Segment> |
| <Segment StartTime="501.284" EndTime="502.798" Participant="mn007"> |
| Tas- task data we are u- |
| </Segment> |
| <Segment StartTime="503.542" EndTime="504.250" Participant="mn007"> |
| Yeah. |
| </Segment> |
| <Segment StartTime="504.532" EndTime="505.562" Participant="me013"> |
| So let me - let me - |
| </Segment> |
| <Segment StartTime="506.082" EndTime="510.202" Participant="me013"> |
| Um, <Pause/> so, <Pause/> um - |
| </Segment> |
| <Segment StartTime="506.914" EndTime="507.476" Participant="mn007"> |
| Mmm. |
| </Segment> |
| <Segment StartTime="510.202" EndTime="513.089" Participant="me013"> |
| <Pause/> O_K, fine. Let's - let's use the conventional meaning of baseline. I - I - |
| </Segment> |
| <Segment StartTime="511.462" EndTime="511.944" Participant="mn007"> |
| Hmm. |
| </Segment> |
| <Segment StartTime="513.089" EndTime="514.409" Participant="me013"> |
| By baseline here I meant |
| </Segment> |
| <Segment StartTime="514.409" EndTime="517.047" Participant="me013"> |
| <Pause/> uh using the task specific data. |
| </Segment> |
| <Segment StartTime="514.863" EndTime="515.457" Participant="mn007"> |
| Oh yeah, the f- |
| </Segment> |
| <Segment StartTime="516.541" EndTime="517.458" Participant="mn007"> |
| Yeah, O_K. Yeah. |
| </Segment> |
| <Segment StartTime="517.047" EndTime="518.488" Participant="me013"> |
| <Pause/> But uh - |
| </Segment> |
| <Segment StartTime="518.488" EndTime="520.620" Participant="me013"> |
| <Pause/> uh, because that's what you were just doing with this ten percent. |
| </Segment> |
| <Segment StartTime="520.620" EndTime="523.007" Participant="me013"> |
| So I was just - I just trying to understand <Emphasis> that. </Emphasis> So if we call |
| </Segment> |
| <Segment StartTime="520.653" EndTime="521.665" Participant="mn007"> |
| Yeah. Sure. |
| </Segment> |
| <Segment StartTime="523.007" EndTime="525.110" Participant="me013"> |
| <Pause/> a factor of w- just <Emphasis> one, </Emphasis> |
| </Segment> |
| <Segment StartTime="525.110" EndTime="527.478" Participant="me013"> |
| just normalized to <Emphasis> one, </Emphasis> the word error rate |
| </Segment> |
| <Segment StartTime="525.110" EndTime="525.624" Participant="mn007"> |
| Mmm. |
| </Segment> |
| <Segment StartTime="525.114" EndTime="526.204" Participant="me006"> |
| <NonVocalSound Description="mike noise"/> |
| </Segment> |
| <Segment StartTime="527.478" EndTime="531.597" Participant="me013"> |
| <Pause/> that you have <Pause/> for using T_I-digits as - as |
| </Segment> |
| <Segment StartTime="531.597" EndTime="534.029" Participant="me013"> |
| <Pause/> <Emphasis> training </Emphasis> and T_I-digits as <Emphasis> test, </Emphasis> |
| </Segment> |
| <Segment StartTime="534.029" EndTime="536.302" Participant="me013"> |
| <Pause/> uh different <Emphasis> words, </Emphasis> I'm sure, but - |
| </Segment> |
| <Segment StartTime="534.326" EndTime="534.748" Participant="mn007"> |
| Mm-hmm. |
| </Segment> |
| <Segment StartTime="536.302" EndTime="539.464" Participant="me013"> |
| <Pause/> but uh, uh the same <Pause/> task and so on. |
| </Segment> |
| <Segment StartTime="539.464" EndTime="542.789" Participant="me013"> |
| <Pause/> If we call that "one", <Pause/> then what you're saying is |
| </Segment> |
| <Segment StartTime="539.493" EndTime="540.265" Participant="mn007"> |
| Mm-hmm. |
| </Segment> |
| <Segment StartTime="542.789" EndTime="549.781" Participant="me013"> |
| <Pause/> that the word error rate <Pause/> for the same language but using <Pause/> uh <Emphasis> different </Emphasis> training data than you're testing on, say TIMIT and so forth, |
| </Segment> |
| <Segment StartTime="549.736" EndTime="550.313" Participant="mn007"> |
| Mm-hmm. |
| </Segment> |
| <Segment StartTime="549.781" EndTime="551.188" Participant="me013"> |
| <Pause/> it's one point one. |
| </Segment> |
| <Segment StartTime="551.849" EndTime="553.774" Participant="mn007"> |
| Yeah, it's around one point one. Yeah. |
| </Segment> |
| <Segment StartTime="552.955" EndTime="553.476" Participant="me013"> |
| Right. |
| </Segment> |
| <Segment StartTime="553.476" EndTime="554.379" Participant="me013"> |
| And if it's - |
| </Segment> |
| <Segment StartTime="554.379" EndTime="556.060" Participant="me013"> |
| <Pause/> you <Pause/> do <Pause/> go to |
| </Segment> |
| <Segment StartTime="556.060" EndTime="558.590" Participant="me013"> |
| <Pause/> three languages <Emphasis> including </Emphasis> the English, |
| </Segment> |
| <Segment StartTime="558.590" EndTime="560.158" Participant="me013"> |
| <Pause/> it's something like one point <Emphasis> three. </Emphasis> |
| </Segment> |
| <Segment StartTime="561.480" EndTime="562.606" Participant="mn007"> |
| Ye- |
| </Segment> |
| <Segment StartTime="561.899" EndTime="563.757" Participant="me013"> |
| <Emphasis> That's </Emphasis> what you were just <Emphasis> saying, </Emphasis> I think. |
| </Segment> |
| <Segment StartTime="563.757" EndTime="566.571" Participant="mn007"> |
| Uh, <Emphasis> more </Emphasis> actually. If I - |
| </Segment> |
| <Segment StartTime="565.212" EndTime="566.351" Participant="me018"> |
| One point four? |
| </Segment> |
| <Segment StartTime="566.571" EndTime="567.171" Participant="mn007"> |
| Yeah. <Comment Description="answers"/> |
| </Segment> |
| <Segment StartTime="569.421" EndTime="571.182" Participant="me018"> |
| So, it's an <Emphasis> additional </Emphasis> thirty percent. |
| </Segment> |
| <Segment StartTime="571.002" EndTime="573.205" Participant="mn007"> |
| What would you say? <Comment Description="probably to [fn002]"/> Around one point four yeah. |
| </Segment> |
| <Segment StartTime="573.029" EndTime="573.553" Participant="me013"> |
| O_K. |
| </Segment> |
| <Segment StartTime="573.553" EndTime="576.101" Participant="me013"> |
| <Pause/> And if you <Emphasis> exclude </Emphasis> <Pause/> English, |
| </Segment> |
| <Segment StartTime="576.101" EndTime="577.829" Participant="me013"> |
| <Pause/> from this combination, what's <Emphasis> that? </Emphasis> |
| </Segment> |
| <Segment StartTime="578.087" EndTime="586.309" Participant="mn007"> |
| If we <Emphasis> exclude </Emphasis> <Emphasis> English, </Emphasis> <Pause/> um <Pause/> there is <Pause/> not much difference with the <Pause/> data <Emphasis> with </Emphasis> English. |
| </Segment> |
| <Segment StartTime="586.553" EndTime="587.149" Participant="me013"> |
| Aha! |
| </Segment> |
| <Segment StartTime="586.554" EndTime="586.862" Participant="mn007"> |
| So. |
| </Segment> |
| <Segment StartTime="587.245" EndTime="587.494" Participant="mn007"> |
| Yeah. |
| </Segment> |
| <Segment StartTime="587.547" EndTime="588.197" Participant="me013"> |
| That's <Emphasis> interesting. </Emphasis> |
| </Segment> |
| <Segment StartTime="588.197" EndTime="589.247" Participant="me013"> |
| <Pause/> That's interesting. |
| </Segment> |
| <Segment StartTime="588.605" EndTime="589.247" Participant="mn007"> |
| <VocalSound Description="laugh"/> |
| </Segment> |
| <Segment StartTime="589.247" EndTime="590.797" Participant="me013"> |
| <Pause/> Do you see? Because - |
| </Segment> |
| <Segment StartTime="589.988" EndTime="590.656" Participant="mn007"> |
| Uh. |
| </Segment> |
| <Segment StartTime="590.797" EndTime="591.578" Participant="me013"> |
| Uh, so - |
| </Segment> |
| <Segment StartTime="591.578" EndTime="596.030" Participant="me013"> |
| No, that - that's important. So what - what it's saying here is just that "yes, there is a reduction |
| </Segment> |
| <Segment StartTime="596.030" EndTime="597.440" Participant="me013"> |
| <Pause/> in performance, |
| </Segment> |
| <Segment StartTime="597.440" EndTime="599.301" Participant="me013"> |
| <Pause/> when you don't <Pause/> um |
| </Segment> |
| <Segment StartTime="599.301" EndTime="600.403" Participant="me013"> |
| <Pause/> have the s- |
| </Segment> |
| <Segment StartTime="600.403" EndTime="603.040" Participant="me013"> |
| <Pause/> when you don't have <Pause/> um |
| </Segment> |
| <Segment StartTime="604.125" EndTime="605.120" Participant="me018"> |
| <Emphasis> Task </Emphasis> data. <Comment Description="completing [me013]'s utterance"/> |
| </Segment> |
| <Segment StartTime="605.687" EndTime="606.952" Participant="me013"> |
| <VocalSound Description="outbreath"/> <VocalSound Description="inbreath"/> |
| </Segment> |
| <Segment StartTime="606.952" EndTime="608.484" Participant="me013"> |
| Wait a minute, th- th- the - |
| </Segment> |
| <Segment StartTime="611.115" EndTime="611.708" Participant="mn007"> |
| Hmm. |
| </Segment> |
| <Segment StartTime="612.205" EndTime="612.609" Participant="me013"> |
| <VocalSound Description="inbreath"/> |
| </Segment> |
| <Segment StartTime="612.609" EndTime="614.900" Participant="me013"> |
| No, actually <Pause/> it's <Emphasis> interesting. </Emphasis> So it's - |
| </Segment> |
| <Segment StartTime="614.900" EndTime="618.711" Participant="me013"> |
| So when you go to a different <Emphasis> task, </Emphasis> there's actually not so <Pause/> different. It's when you went to these - |
| </Segment> |
| <Segment StartTime="618.711" EndTime="620.802" Participant="me013"> |
| So what's the difference between two and three? |
| </Segment> |
| <Segment StartTime="620.802" EndTime="623.783" Participant="me013"> |
| <Pause/> Between the one point one case and the one point four case? I'm confused. |
| </Segment> |
| <Segment StartTime="626.658" EndTime="627.743" Participant="me018"> |
| It's multilingual. |
| </Segment> |
| <Segment StartTime="629.237" EndTime="632.148" Participant="mn007"> |
| Yeah. The only difference it's - is that it's multilingual - |
| </Segment> |
| <Segment StartTime="632.987" EndTime="635.482" Participant="mn007"> |
| Um <Pause/> Yeah. |
| </Segment> |
| <Segment StartTime="633.635" EndTime="637.354" Participant="me013"> |
| Cuz in both - in both - <Emphasis> both </Emphasis> of those cases, you don't have the same <Emphasis> task. </Emphasis> |
| </Segment> |
| <Segment StartTime="637.867" EndTime="638.861" Participant="mn007"> |
| Yeah sure. |
| </Segment> |
| <Segment StartTime="638.474" EndTime="642.448" Participant="me013"> |
| <Pause/> So is - is the training data for the - for this one point four case - |
| </Segment> |
| <Segment StartTime="642.448" EndTime="646.420" Participant="me013"> |
| <Pause/> does it <Emphasis> include </Emphasis> the training data for the one point one case? |
| </Segment> |
| <Segment StartTime="647.256" EndTime="649.594" Participant="mn007"> |
| Uh yeah. |
| </Segment> |
| <Segment StartTime="648.163" EndTime="649.774" Participant="me006"> |
| Yeah, a fraction of it. |
| </Segment> |
| <Segment StartTime="649.594" EndTime="651.299" Participant="mn007"> |
| <Pause/> A part of it, yeah. |
| </Segment> |
| <Segment StartTime="651.814" EndTime="652.844" Participant="me006"> |
| <VocalSound Description="clear throat"/> |
| </Segment> |
| <Segment StartTime="654.979" EndTime="656.420" Participant="me013"> |
| How m- how much bigger is it? |
| </Segment> |
| <Segment StartTime="658.770" EndTime="660.826" Participant="mn007"> |
| Um <Pause/> |
| </Segment> |
| <Segment StartTime="661.752" EndTime="662.887" Participant="me006"> |
| Yeah, um. |
| </Segment> |
| <Segment StartTime="662.218" EndTime="664.317" Participant="mn007"> |
| It's two times, actually? Yeah. |
| </Segment> |
| <Segment StartTime="665.453" EndTime="667.010" Participant="mn007"> |
| Um. |
| </Segment> |
| <Segment StartTime="667.447" EndTime="670.682" Participant="mn007"> |
| The <Emphasis> English </Emphasis> data - |
| </Segment> |
| <Segment StartTime="670.682" EndTime="676.551" Participant="mn007"> |
| <Pause/> No, the <Emphasis> multilingual </Emphasis> databases are two times the <Pause/> broad English <Pause/> data. |
| </Segment> |
| <Segment StartTime="677.935" EndTime="681.545" Participant="mn007"> |
| We just wanted to keep this, w- well, not too <Emphasis> huge. </Emphasis> So. |
| </Segment> |
| <Segment StartTime="682.011" EndTime="684.206" Participant="me013"> |
| So it's <Emphasis> two </Emphasis> times, |
| </Segment> |
| <Segment StartTime="684.206" EndTime="687.112" Participant="me013"> |
| <Pause/> but it <Emphasis> includes </Emphasis> the - but it <Emphasis> includes </Emphasis> the broad English data. |
| </Segment> |
| <Segment StartTime="684.808" EndTime="685.618" Participant="mn007"> |
| I think so. <Uncertain> Do you </Uncertain> - <Comment Description="probably to [fn002]"/> |
| </Segment> |
| <Segment StartTime="687.561" EndTime="688.296" Participant="mn007"> |
| Uh, |
| </Segment> |
| <Segment StartTime="688.946" EndTime="689.436" Participant="mn007"> |
| Yeah. |
| </Segment> |
| <Segment StartTime="689.792" EndTime="694.071" Participant="me013"> |
| And the broad English data is what you got this one point one <Pause/> with. So that's TIMIT basically right? |
| </Segment> |
| <Segment StartTime="694.013" EndTime="694.483" Participant="me006"> |
| Mm-hmm. |
| </Segment> |
| <Segment StartTime="694.035" EndTime="694.494" Participant="mn007"> |
| Yeah. |
| </Segment> |
| <Segment StartTime="694.631" EndTime="696.037" Participant="me013"> |
| So it's band-limited TIMIT. |
| </Segment> |
| <Segment StartTime="696.395" EndTime="697.102" Participant="me006"> |
| Mm-hmm. |
| </Segment> |
| <Segment StartTime="696.909" EndTime="697.358" Participant="mn007"> |
| Mm-hmm. |
| </Segment> |
| <Segment StartTime="696.928" EndTime="697.810" Participant="me013"> |
| This is all |
| </Segment> |
| <Segment StartTime="697.864" EndTime="698.216" Participant="mn007"> |
| Yeah. |
| </Segment> |
| <Segment StartTime="697.967" EndTime="698.631" Participant="me006"> |
| <VocalSound Description="mouth"/> <VocalSound Description="inbreath"/> |
| </Segment> |
| <Segment StartTime="698.631" EndTime="699.777" Participant="me006"> |
| Downs- |
| </Segment> |
| <Segment StartTime="698.804" EndTime="699.834" Participant="me013"> |
| eight kilohertz sampling. |
| </Segment> |
| <Segment StartTime="699.777" EndTime="700.508" Participant="me006"> |
| Right. |
| </Segment> |
| <Segment StartTime="700.802" EndTime="703.220" Participant="me013"> |
| So you have band-limited TIMIT, <Pause/> gave you uh |
| </Segment> |
| <Segment StartTime="702.277" EndTime="703.822" Participant="me006"> |
| <VocalSound Description="clear throat"/> |
| </Segment> |
| <Segment StartTime="703.220" EndTime="705.768" Participant="me013"> |
| <Emphasis> almost </Emphasis> as good as a result as using T_I-<Emphasis> digits </Emphasis> |
| </Segment> |
| <Segment StartTime="705.768" EndTime="707.436" Participant="me013"> |
| <Pause/> on a T_I-digits <Emphasis> test. </Emphasis> |
| </Segment> |
| <Segment StartTime="708.229" EndTime="708.717" Participant="me013"> |
| O_K? |
| </Segment> |
| <Segment StartTime="708.575" EndTime="709.457" Participant="mn007"> |
| Hmm? |
| </Segment> |
| <Segment StartTime="709.566" EndTime="712.146" Participant="me013"> |
| Um <Pause/> and <Pause/> um |
| </Segment> |
| <Segment StartTime="712.146" EndTime="713.320" Participant="me013"> |
| <Pause/> But, |
| </Segment> |
| <Segment StartTime="713.320" EndTime="717.874" Participant="me013"> |
| <Pause/> when you add in more training data but keep the neural net the same size, |
| </Segment> |
| <Segment StartTime="717.874" EndTime="722.239" Participant="me013"> |
| <Pause/> it <Pause/> um performs worse on the T_I-digits. |
| </Segment> |
| <Segment StartTime="722.239" EndTime="724.859" Participant="me013"> |
| <Pause/> O_K, now all of this is - |
| </Segment> |
| <Segment StartTime="724.859" EndTime="727.226" Participant="me013"> |
| <Pause/> This is <Emphasis> noisy </Emphasis> <Pause/> T_I-digits, I assume? |
| </Segment> |
| <Segment StartTime="727.455" EndTime="728.154" Participant="mn007"> |
| Yep. |
| </Segment> |
| <Segment StartTime="727.750" EndTime="728.924" Participant="me013"> |
| Both training and test? |
| </Segment> |
| <Segment StartTime="729.593" EndTime="730.298" Participant="me013"> |
| Yeah. <Comment Description="answers question"/> O_K. |
| </Segment> |
| <Segment StartTime="731.538" EndTime="732.920" Participant="me013"> |
| Um |
| </Segment> |
| <Segment StartTime="736.397" EndTime="736.908" Participant="me013"> |
| O_K. |
| </Segment> |
| <Segment StartTime="737.278" EndTime="737.631" Participant="me013"> |
| Well. |
| </Segment> |
| <Segment StartTime="737.631" EndTime="741.263" Participant="me013"> |
| <Pause/> We - we - we may just need to uh - |
| </Segment> |
| <Segment StartTime="741.809" EndTime="743.869" Participant="me013"> |
| So I mean it's interesting that h- going to a different - |
| </Segment> |
| <Segment StartTime="743.869" EndTime="747.953" Participant="me013"> |
| different <Emphasis> task </Emphasis> didn't seem to <Emphasis> hurt </Emphasis> us that much, and going to a different <Emphasis> language </Emphasis> |
| </Segment> |
| <Segment StartTime="749.001" EndTime="750.400" Participant="me013"> |
| um |
| </Segment> |
| <Segment StartTime="752.682" EndTime="753.877" Participant="me013"> |
| It doesn't seem to matter - |
| </Segment> |
| <Segment StartTime="753.877" EndTime="757.581" Participant="me013"> |
| <Pause/> The difference between <Emphasis> three </Emphasis> and <Emphasis> four </Emphasis> is not particularly great, so that means that |
| </Segment> |
| <Segment StartTime="757.581" EndTime="761.022" Participant="me013"> |
| <Pause/> whether you have the <Emphasis> language </Emphasis> in or <Emphasis> not </Emphasis> is not such a big deal. |
| </Segment> |
| <Segment StartTime="761.022" EndTime="762.186" Participant="mn007"> |
| Mmm. |
| </Segment> |
| <Segment StartTime="762.186" EndTime="763.759" Participant="me013"> |
| It sounds like um |
| </Segment> |
| <Segment StartTime="763.759" EndTime="765.621" Participant="me013"> |
| <Pause/> uh |
| </Segment> |
| <Segment StartTime="765.621" EndTime="768.201" Participant="me013"> |
| <Pause/> we <Emphasis> may </Emphasis> need to have more |
| </Segment> |
| <Segment StartTime="768.201" EndTime="773.568" Participant="me013"> |
| <Pause/> of uh things that are similar to a target language or - I mean. |
| </Segment> |
| <Segment StartTime="773.568" EndTime="778.664" Participant="me013"> |
| <Pause/> You have the <Emphasis> same </Emphasis> number of parameters in the <Emphasis> neural </Emphasis> net, you haven't increased the size of the neural net, |
| </Segment> |
| <Segment StartTime="778.664" EndTime="780.142" Participant="me013"> |
| <Pause/> and maybe there's just - |
| </Segment> |
| <Segment StartTime="780.142" EndTime="781.450" Participant="me013"> |
| <Pause/> just not enough |
| </Segment> |
| <Segment StartTime="781.450" EndTime="783.818" Participant="me013"> |
| <Pause/> complexity to it to represent |
| </Segment> |
| <Segment StartTime="783.818" EndTime="785.751" Participant="me013"> |
| <Pause/> the variab- increased variability in the - |
| </Segment> |
| <Segment StartTime="785.751" EndTime="786.781" Participant="me013"> |
| in the training set. |
| </Segment> |
| <Segment StartTime="786.781" EndTime="787.829" Participant="me013"> |
| <Pause/> That - that could be. |
| </Segment> |
| <Segment StartTime="789.454" EndTime="790.059" Participant="me013"> |
| Um |
| </Segment> |
| <Segment StartTime="790.059" EndTime="790.998" Participant="me013"> |
| <Pause/> So, what about - |
| </Segment> |
| <Segment StartTime="790.998" EndTime="792.841" Participant="me013"> |
| So <Emphasis> these </Emphasis> are results with |
| </Segment> |
| <Segment StartTime="792.841" EndTime="793.456" Participant="me013"> |
| <Pause/> uh th- |
| </Segment> |
| <Segment StartTime="793.456" EndTime="794.739" Participant="me013"> |
| <Pause/> that you're describing now, that |
| </Segment> |
| <Segment StartTime="794.739" EndTime="797.612" Participant="me013"> |
| <Pause/> they are pretty similar for the different features or - |
| </Segment> |
| <Segment StartTime="797.612" EndTime="798.584" Participant="me013"> |
| <Pause/> or uh - |
| </Segment> |
| <Segment StartTime="798.584" EndTime="800.481" Participant="mn007"> |
| Uh, let me check. Uh. <VocalSound Description="laugh"/> |
| </Segment> |
| <Segment StartTime="800.481" EndTime="803.607" Participant="mn007"> |
| <Pause/> So. This was for the P_L_P, |
| </Segment> |
| <Segment StartTime="800.709" EndTime="801.154" Participant="me013"> |
| Yeah. |
| </Segment> |
| <Segment StartTime="803.607" EndTime="809.209" Participant="mn007"> |
| <Pause/> Um. The - Yeah. For the P_L_P with J_RASTA the - |
| </Segment> |
| <Segment StartTime="803.624" EndTime="804.150" Participant="me013"> |
| Yeah. |
| </Segment> |
| <Segment StartTime="809.209" EndTime="811.185" Participant="mn007"> |
| <Pause/> the - we - |
| </Segment> |
| <Segment StartTime="811.185" EndTime="814.380" Participant="mn007"> |
| <Pause/> This is quite the same <Pause/> tendency, |
| </Segment> |
| <Segment StartTime="814.380" EndTime="817.344" Participant="mn007"> |
| <Pause/> with a slight increase of the error rate, |
| </Segment> |
| <Segment StartTime="817.344" EndTime="820.069" Participant="mn007"> |
| <Pause/> uh if we go to - to TIMIT. |
| </Segment> |
| <Segment StartTime="820.069" EndTime="824.333" Participant="mn007"> |
| And then it's - it gets worse with the multilingual. |
| </Segment> |
| <Segment StartTime="824.333" EndTime="826.722" Participant="mn007"> |
| <Pause/> Um. |
| </Segment> |
| <Segment StartTime="828.309" EndTime="833.079" Participant="mn007"> |
| Yeah. There - there is a difference actually with - b- between P_L_P and J_RASTA is that |
| </Segment> |
| <Segment StartTime="833.079" EndTime="839.603" Participant="mn007"> |
| <Pause/> J_RASTA <Pause/> seems to <Pause/> perform better with the highly mismatched <Pause/> condition |
| </Segment> |
| <Segment StartTime="839.603" EndTime="844.687" Participant="mn007"> |
| <Pause/> but <Pronounce> slightly </Pronounce> - slightly worse <Pause/> for the well matched condition. |
| </Segment> |
| <Segment StartTime="845.881" EndTime="846.721" Participant="mn007"> |
| Mmm. |
| </Segment> |
| <Segment StartTime="846.530" EndTime="846.852" Participant="me013"> |
| <VocalSound Description="inbreath"/> |
| </Segment> |
| <Segment StartTime="846.852" EndTime="849.801" Participant="me013"> |
| I have a suggestion, actually, even though it'll delay us slightly, would - would you mind |
| </Segment> |
| <Segment StartTime="849.801" EndTime="851.402" Participant="me013"> |
| <Pause/> running into the other room and making |
| </Segment> |
| <Segment StartTime="851.402" EndTime="854.239" Participant="me013"> |
| <Pause/> copies of this? <Pause/> Cuz we're all sort of - |
| </Segment> |
| <Segment StartTime="853.776" EndTime="854.977" Participant="mn007"> |
| Yeah, yeah. |
| </Segment> |
| <Segment StartTime="854.239" EndTime="856.830" Participant="me013"> |
| <Pause/> If we c- if we could look at it, while we're talking, I think <Uncertain> it </Uncertain>'d be uh - |
| </Segment> |
| <Segment StartTime="854.977" EndTime="856.894" Participant="mn007"> |
| <Pause/> O_K. |
| </Segment> |
| <Segment StartTime="856.830" EndTime="863.105" Participant="me013"> |
| <Pause/> Uh, I'll - I'll sing a song or dance or something while you <VocalSound Description="laugh"/> do it, too. <VocalSound Description="laugh"/> |
| </Segment> |
| <Segment StartTime="858.218" EndTime="861.845" Participant="mn007"> |
| <NonVocalSound Description="mike noise"/> |
| </Segment> |
| <Segment StartTime="860.698" EndTime="863.150" Participant="me006"> |
| Alright. <NonVocalSound Description="mike noise"/> |
| </Segment> |
| <Segment StartTime="860.816" EndTime="861.311" Participant="me018"> |
| <VocalSound Description="mouth"/> |
| </Segment> |
| <Segment StartTime="861.311" EndTime="866.349" Participant="me018"> |
| So um - Go ahead. Ah, while you're gone I'll ask s- some of my questions. <VocalSound Description="laugh"/> |
| </Segment> |
| <Segment StartTime="863.611" EndTime="864.135" Participant="me013"> |
| Yeah. |
| </Segment> |
| <Segment StartTime="866.196" EndTime="867.816" Participant="me006"> |
| <NonVocalSound Description="mike noise"/> |
| </Segment> |
| <Segment StartTime="866.349" EndTime="867.275" Participant="me018"> |
| Um. |
| </Segment> |
| <Segment StartTime="866.418" EndTime="872.829" Participant="mn007"> |
| <Comment Description="off-microphone, [mn007] asks where the copy machine is"/> |
| </Segment> |
| <Segment StartTime="867.371" EndTime="868.280" Participant="me013"> |
| Yeah. <VocalSound Description="laugh"/> |
| </Segment> |
| <Segment StartTime="868.950" EndTime="869.896" Participant="me013"> |
| Uh, |
| </Segment> |
| <Segment StartTime="870.347" EndTime="872.299" Participant="me013"> |
| this way and just slightly to the left, yeah. |
| </Segment> |
| <Segment StartTime="873.836" EndTime="875.333" Participant="me018"> |
| The um - |
| </Segment> |
| <Segment StartTime="875.921" EndTime="877.140" CloseMic="false"> |
| <NonVocalSound Description="door slams"/> |
| </Segment> |
| <Segment StartTime="876.376" EndTime="884.127" Participant="me018"> |
| What was - Was this number <Pause/> forty or - It was roughly the same as this one, <Pause/> he said? <Pause/> When you had the <Emphasis> two </Emphasis> language versus the <Emphasis> three </Emphasis> language? |
| </Segment> |
| <Segment StartTime="881.379" EndTime="882.225" Participant="me013"> |
| Um. |
| </Segment> |
| <Segment StartTime="883.969" EndTime="884.959" Participant="me006"> |
| <VocalSound Description="outbreath"/> |
| </Segment> |
| <Segment StartTime="883.970" EndTime="885.022" Participant="me013"> |
| That's what he was <Emphasis> saying. </Emphasis> |
| </Segment> |
| <Segment StartTime="884.940" EndTime="887.208" Participant="me018"> |
| That's where he removed English, right? |
| </Segment> |
| <Segment StartTime="884.959" EndTime="885.718" Participant="me006"> |
| Yeah. |
| </Segment> |
| <Segment StartTime="885.718" EndTime="886.641" Participant="me006"> |
| <VocalSound Description="clear throat"/> |
| </Segment> |
| <Segment StartTime="886.561" EndTime="887.208" Participant="me013"> |
| Right. |
| </Segment> |
| <Segment StartTime="886.641" EndTime="887.208" Participant="me006"> |
| <VocalSound Description="inbreath"/> |
| </Segment> |
| <Segment StartTime="887.208" EndTime="890.430" Participant="me006"> |
| It sometimes, actually, depends on what features you're using. |
| </Segment> |
| <Segment StartTime="890.615" EndTime="891.576" Participant="me013"> |
| Yeah. |
| </Segment> |
| <Segment StartTime="891.576" EndTime="893.882" Participant="me013"> |
| But - but i- it sounds like - |
| </Segment> |
| <Segment StartTime="891.956" EndTime="892.516" Participant="me006"> |
| Um, but - |
| </Segment> |
| <Segment StartTime="893.882" EndTime="895.956" Participant="me013"> |
| <Pause/> I mean. <Emphasis> That's </Emphasis> interesting because |
| </Segment> |
| <Segment StartTime="895.956" EndTime="899.719" Participant="me013"> |
| <Pause/> it - it <Emphasis> seems </Emphasis> like what it's saying is <Emphasis> not </Emphasis> so much that you got <Emphasis> hurt </Emphasis> |
| </Segment> |
| <Segment StartTime="899.719" EndTime="902.731" Participant="me013"> |
| <Pause/> uh because <Pause/> you |
| </Segment> |
| <Segment StartTime="902.731" EndTime="907.227" Participant="me013"> |
| <Pause/> uh didn't have so much representation of <Emphasis> English, </Emphasis> |
| </Segment> |
| <Segment StartTime="907.227" EndTime="911.379" Participant="me013"> |
| <Pause/> because in the <Emphasis> other </Emphasis> case you don't get hurt <Emphasis> any </Emphasis> <Emphasis> more, </Emphasis> <Pause/> at least when |
| </Segment> |
| <Segment StartTime="911.379" EndTime="917.359" Participant="me013"> |
| <Pause/> it seemed like uh it - it might simply be a case that you <Emphasis> have </Emphasis> something that is just much more <Emphasis> diverse, </Emphasis> |
| </Segment> |
| <Segment StartTime="917.218" EndTime="918.032" Participant="me018"> |
| Mm-hmm. |
| </Segment> |
| <Segment StartTime="917.359" EndTime="920.142" Participant="me013"> |
| <Pause/> but you have the same number of parameters <Emphasis> representing </Emphasis> it. |
| </Segment> |
| <Segment StartTime="919.929" EndTime="931.338" Participant="me018"> |
| Mm-hmm. <Pause/> I wonder - were um all three of these nets <Pause/> using the same output? This <Pronounce Pronunciation=""multi..language""> multi-language </Pronounce> <Pause/> uh labeling? |
| </Segment> |
| <Segment StartTime="927.735" EndTime="928.869" Participant="me006"> |
| <VocalSound Description="mouth"/> <VocalSound Description="inbreath"/> He - |
| </Segment> |
| <Segment StartTime="929.413" EndTime="930.156" Participant="me006"> |
| Mm-hmm. |
| </Segment> |
| <Segment StartTime="931.166" EndTime="934.798" Participant="me006"> |
| He was using uh sixty-four phonemes from <Pause/> SAMPA. |
| </Segment> |
| <Segment StartTime="934.779" EndTime="936.273" Participant="me018"> |
| O_K, O_K. |
| </Segment> |
| <Segment StartTime="935.358" EndTime="935.955" Participant="me006"> |
| Yeah. |
| </Segment> |
| <Segment StartTime="939.215" EndTime="939.703" Participant="me018"> |
| <VocalSound Description="inbreath"/> |
| </Segment> |
| <Segment StartTime="939.703" EndTime="940.425" Participant="me018"> |
| So <Emphasis> this </Emphasis> would - |
| </Segment> |
| <Segment StartTime="940.425" EndTime="944.006" Participant="me018"> |
| <Pause/> From <Emphasis> this </Emphasis> you would say, "well, it doesn't really <Emphasis> matter </Emphasis> if we put Finnish |
| </Segment> |
| <Segment StartTime="944.006" EndTime="947.078" Participant="me018"> |
| <Pause/> into <Pause/> the training of the neural net, |
| </Segment> |
| <Segment StartTime="947.078" EndTime="948.957" Participant="me018"> |
| <Pause/> if there's <Pause/> gonna be, |
| </Segment> |
| <Segment StartTime="948.957" EndTime="951.311" Participant="me018"> |
| <Pause/> you know, Finnish in the test data." Right? |
| </Segment> |
| <Segment StartTime="951.530" EndTime="953.060" Participant="me013"> |
| Well, it's - it sounds - |
| </Segment> |
| <Segment StartTime="953.060" EndTime="956.128" Participant="me013"> |
| <Pause/> I mean, we have to be <Emphasis> careful, </Emphasis> cuz we haven't gotten a <Emphasis> good </Emphasis> result yet. |
| </Segment> |
| <Segment StartTime="956.128" EndTime="957.065" Participant="me013"> |
| <VocalSound Description="laugh, inbreath"/> |
| </Segment> |
| <Segment StartTime="956.309" EndTime="957.065" Participant="me018"> |
| Yeah. |
| </Segment> |
| <Segment StartTime="957.065" EndTime="959.342" Participant="me013"> |
| And comparing different <Emphasis> bad </Emphasis> results can be <Pause/> tricky. |
| </Segment> |
| <Segment StartTime="958.703" EndTime="959.499" Participant="me018"> |
| Hmm. |
| </Segment> |
| <Segment StartTime="959.342" EndTime="960.798" Participant="me013"> |
| <Pause/> But I - I - I - |
| </Segment> |
| <Segment StartTime="960.798" EndTime="964.701" Participant="me013"> |
| <Pause/> I think it <Emphasis> does </Emphasis> suggest that it's not so much uh |
| </Segment> |
| <Segment StartTime="964.701" EndTime="971.845" Participant="me013"> |
| <Pause/> uh cross <Pause/> <Emphasis> language </Emphasis> as cross <Emphasis> type </Emphasis> of <Emphasis> speech. </Emphasis> |
| </Segment> |
| <Segment StartTime="972.340" EndTime="973.123" Participant="me018"> |
| Mm-hmm. |
| </Segment> |
| <Segment StartTime="976.114" EndTime="978.373" Participant="me013"> |
| It's - it's um - |
| </Segment> |
| <Segment StartTime="983.590" EndTime="983.969" Participant="me013"> |
| <VocalSound Description="inbreath"/> |
| </Segment> |
| <Segment StartTime="983.969" EndTime="987.748" Participant="me013"> |
| But we <Emphasis> did </Emphasis> - Oh yeah, the other thing I was <Emphasis> asking </Emphasis> him, though, is that I <Emphasis> think </Emphasis> that in the case - |
| </Segment> |
| <Segment StartTime="987.748" EndTime="992.645" Participant="me013"> |
| <Pause/> Yeah, you - you <Emphasis> do </Emphasis> have to be careful because of com- compounded results. I think we got some earlier results |
| </Segment> |
| <Segment StartTime="992.645" EndTime="996.689" Participant="me013"> |
| <Pause/> in which you <Emphasis> trained </Emphasis> on one language and <Emphasis> tested </Emphasis> on <Emphasis> another </Emphasis> and you didn't have |
| </Segment> |
| <Segment StartTime="996.689" EndTime="1000.086" Participant="me013"> |
| <Pause/> <Emphasis> three, </Emphasis> but you just had <Emphasis> one </Emphasis> <Pause/> language. So you trained on |
| </Segment> |
| <Segment StartTime="1000.086" EndTime="1002.493" Participant="me013"> |
| <Pause/> one type of digits and <Emphasis> tested </Emphasis> on <Emphasis> another. </Emphasis> Didn- |
| </Segment> |
| <Segment StartTime="1002.493" EndTime="1003.931" Participant="me013"> |
| Wasn't there something of that? Where you, |
| </Segment> |
| <Segment StartTime="1003.931" EndTime="1008.268" Participant="me013"> |
| <Pause/> say, trained on Spanish and tested on - on T_I-digits, or the other way around? |
| </Segment> |
| <Segment StartTime="1008.268" EndTime="1009.207" Participant="me013"> |
| <Pause/> Something like that? |
| </Segment> |
| <Segment StartTime="1008.870" EndTime="1009.487" Participant="fn002"> |
| No. |
| </Segment> |
| <Segment StartTime="1010.057" EndTime="1011.388" Participant="me013"> |
| I thought there was <Emphasis> something </Emphasis> like that, |
| </Segment> |
| <Segment StartTime="1011.388" EndTime="1013.177" Participant="me013"> |
| <Pause/> that he showed me <Pause/> last week. |
| </Segment> |
| <Segment StartTime="1013.780" EndTime="1014.839" Participant="me013"> |
| We'll have to wait till we get - |
| </Segment> |
| <Segment StartTime="1014.943" EndTime="1016.110" Participant="me018"> |
| Yeah, that would be interesting. |
| </Segment> |
| <Segment StartTime="1016.285" EndTime="1016.845" Participant="me013"> |
| <VocalSound Description="inbreath"/> |
| </Segment> |
| <Segment StartTime="1016.845" EndTime="1017.207" Participant="me013"> |
| Um, |
| </Segment> |
| <Segment StartTime="1018.001" EndTime="1019.616" CloseMic="false"> |
| <NonVocalSound Description="door opens"/> <Comment Description="[mn007] returns"/> |
| </Segment> |
| <Segment StartTime="1018.140" EndTime="1020.256" Participant="me013"> |
| This <Emphasis> may </Emphasis> have been what I was asking before, Stephane, but - |
| </Segment> |
| <Segment StartTime="1019.616" EndTime="1020.799" CloseMic="false"> |
| <NonVocalSound Description="door closes"/> |
| </Segment> |
| <Segment StartTime="1020.256" EndTime="1023.093" Participant="me013"> |
| <Pause/> but, um, wasn't there <Emphasis> something </Emphasis> that you did, |
| </Segment> |
| <Segment StartTime="1023.093" EndTime="1027.050" Participant="me013"> |
| <Pause/> where you <Emphasis> trained </Emphasis> <Pause/> on <Emphasis> one </Emphasis> language and <Emphasis> tested </Emphasis> on <Emphasis> another? </Emphasis> |
| </Segment> |
| <Segment StartTime="1027.773" EndTime="1029.650" Participant="me013"> |
| I mean no - no <Emphasis> mixture </Emphasis> but just - |
| </Segment> |
| <Segment StartTime="1031.473" EndTime="1032.754" Participant="me006"> |
| <Uncertain> I'll get it for you. </Uncertain> <Comment Description="whispered"/> |
| </Segment> |
| <Segment StartTime="1035.337" EndTime="1037.236" Participant="mn007"> |
| Uh, no, no. <Comment Description="perhaps un-miked"/> |
| </Segment> |
| <Segment StartTime="1037.639" EndTime="1039.299" Participant="me013"> |
| We've never just trained on one lang- |
| </Segment> |
| <Segment StartTime="1038.717" EndTime="1042.116" Participant="mn007"> |
| Training on a <Emphasis> single </Emphasis> language, you mean, and <Emphasis> testing </Emphasis> on the <Emphasis> other </Emphasis> one? <Comment Description="perhaps un-miked"/> |
| </Segment> |
| <Segment StartTime="1042.011" EndTime="1042.729" Participant="me013"> |
| Yeah. |
| </Segment> |
| <Segment StartTime="1042.313" EndTime="1044.634" Participant="fn002"> |
| <VocalSound Description="inbreath"/> Not yet. |
| </Segment> |
| <Segment StartTime="1043.264" EndTime="1043.689" Participant="mn007"> |
| Uh, no. |
| </Segment> |
| <Segment StartTime="1044.086" EndTime="1048.611" Participant="mn007"> |
| So the only <Pause/> task that's similar to this is the training on <Emphasis> two </Emphasis> languages, and <Comment Description="perhaps un-miked"/> |
| </Segment> |
| <Segment StartTime="1048.975" EndTime="1051.595" Participant="me013"> |
| But we've <Emphasis> done </Emphasis> a bunch of things where we just trained on one <Emphasis> language. </Emphasis> Right? |
| </Segment> |
| <Segment StartTime="1049.348" EndTime="1049.968" Participant="mn007"> |
| that - |
| </Segment> |
| <Segment StartTime="1052.538" EndTime="1055.866" Participant="me013"> |
| I mean, you haven't - you haven't done <Emphasis> all </Emphasis> your tests on multiple <Emphasis> languages. </Emphasis> |
| </Segment> |
| <Segment StartTime="1056.590" EndTime="1057.482" Participant="mn007"> |
| Uh, |
| </Segment> |
| <Segment StartTime="1058.530" EndTime="1058.963" Participant="mn007"> |
| <Emphasis> No. </Emphasis> |
| </Segment> |
| <Segment StartTime="1058.963" EndTime="1069.227" Participant="mn007"> |
| <Pause/> Either thi- this is test with <Pause/> uh the <Emphasis> same </Emphasis> language <Pause/> but from the <Emphasis> broad </Emphasis> data, or it's test with <Pause/> uh <Emphasis> different </Emphasis> languages <Comment Description="perhaps un-miked"/> |
| </Segment> |
| <Segment StartTime="1069.227" EndTime="1071.652" Participant="mn007"> |
| <Emphasis> also </Emphasis> from the broad data, <Emphasis> excluding </Emphasis> the - <Comment Description="perhaps un-miked"/> |
| </Segment> |
| <Segment StartTime="1071.099" EndTime="1073.446" Participant="fn002"> |
| The <Uncertain> early </Uncertain> experiment that - |
| </Segment> |
| <Segment StartTime="1071.652" EndTime="1073.394" Participant="mn007"> |
| So, it's - it's three or - three and four. <Comment Description="perhaps un-miked"/> |
| </Segment> |
| <Segment StartTime="1072.218" EndTime="1075.007" Participant="me018"> |
| Did you do different languages from digits? |
| </Segment> |
| <Segment StartTime="1077.248" EndTime="1078.459" Participant="mn007"> |
| Uh. No. |
| </Segment> |
| <Segment StartTime="1078.459" EndTime="1079.564" Participant="mn007"> |
| You mean |
| </Segment> |
| <Segment StartTime="1079.564" EndTime="1081.305" Participant="mn007"> |
| <Pause/> training digits |
| </Segment> |
| <Segment StartTime="1081.305" EndTime="1085.060" Participant="mn007"> |
| <Pause/> on one language and using the net |
| </Segment> |
| <Segment StartTime="1085.060" EndTime="1087.461" Participant="mn007"> |
| <Pause/> to recognize on the other? |
| </Segment> |
| <Segment StartTime="1086.654" EndTime="1088.958" Participant="me018"> |
| Digits on another language? <Comment Description="completing [mn007]'s utterance"/> |
| </Segment> |
| <Segment StartTime="1088.108" EndTime="1088.433" Participant="mn007"> |
| No. <Comment Description="mike is back on"/> |
| </Segment> |
| <Segment StartTime="1091.229" EndTime="1093.531" Participant="me013"> |
| See, I <Emphasis> thought </Emphasis> you <Emphasis> showed </Emphasis> me something like that last <Emphasis> week. </Emphasis> |
| </Segment> |
| <Segment StartTime="1093.531" EndTime="1094.846" Participant="me013"> |
| You had a - you had a little - |
| </Segment> |
| <Segment StartTime="1094.937" EndTime="1097.304" Participant="mn007"> |
| Uh, <Pause/> |
| </Segment> |
| <Segment StartTime="1097.304" EndTime="1098.478" Participant="mn007"> |
| No, I don't <Emphasis> think </Emphasis> so. |
| </Segment> |
| <Segment StartTime="1103.404" EndTime="1104.384" Participant="me013"> |
| Um |
| </Segment> |
| <Segment StartTime="1106.261" EndTime="1107.039" Participant="me013"> |
| What - |
| </Segment> |
| <Segment StartTime="1109.250" EndTime="1110.569" Participant="me034"> |
| These numbers are uh |
| </Segment> |
| <Segment StartTime="1110.569" EndTime="1111.779" Participant="me034"> |
| <Pause/> ratio to baseline? |
| </Segment> |
| <Segment StartTime="1110.614" EndTime="1116.922" Participant="me013"> |
| So, I mean wha- what's the - This - this chart - this table that we're looking at <Pause/> is um, |
| </Segment> |
| <Segment StartTime="1111.134" EndTime="1111.618" Participant="mn007"> |
| So. |
| </Segment> |
| <Segment StartTime="1119.723" EndTime="1122.078" Participant="me013"> |
| show- is all <Emphasis> testing </Emphasis> for <Emphasis> T_I </Emphasis>-digits, or - ? <Comment Description="leading question"/> |
| </Segment> |
| <Segment StartTime="1122.169" EndTime="1125.490" Participant="me006"> |
| Bigger is worse. <Pause/> This is error rate, I think. <Comment Description="probably to [me034]"/> |
| </Segment> |
| <Segment StartTime="1123.116" EndTime="1128.903" Participant="mn007"> |
| So you have uh basically two <Pause/> uh parts. The upper part is for T_I-digits |
| </Segment> |
| <Segment StartTime="1125.440" EndTime="1126.313" Participant="me034"> |
| <Emphasis> Ratio. </Emphasis> <Comment Description="probably to [me006]"/> |
| </Segment> |
| <Segment StartTime="1125.490" EndTime="1126.954" Participant="me006"> |
| No. <Pause/> No. <Comment Description="probably to [me034], correcting himself"/> |
| </Segment> |
| <Segment StartTime="1127.737" EndTime="1128.729" Participant="me034"> |
| <VocalSound Description="mouth"/> <VocalSound Description="inbreath"/> |
| </Segment> |
| <Segment StartTime="1127.984" EndTime="1129.507" Participant="me006"> |
| Yeah, yeah, yeah. <Comment Description="probably to [me034], probably agreeing with [me034]"/> |
| </Segment> |
| <Segment StartTime="1128.903" EndTime="1134.324" Participant="mn007"> |
| <Pause/> and it's divided in three <Pause/> rows <Pause/> of four - four rows each. |
| </Segment> |
| <Segment StartTime="1130.459" EndTime="1130.947" Participant="me006"> |
| Mm-hmm. |
| </Segment> |
| <Segment StartTime="1134.980" EndTime="1135.674" Participant="me013"> |
| Yeah. |
| </Segment> |
| <Segment StartTime="1135.284" EndTime="1141.822" Participant="mn007"> |
| And the first four rows is well-matched, then the s- the second group of four rows is mismatched, and |
| </Segment> |
| <Segment StartTime="1141.822" EndTime="1144.198" Participant="mn007"> |
| <Pause/> finally highly mismatched. |
| </Segment> |
| <Segment StartTime="1144.198" EndTime="1147.820" Participant="mn007"> |
| <Pause/> And then the lower part is for Italian and it's the same - |
| </Segment> |
| <Segment StartTime="1147.820" EndTime="1149.804" Participant="mn007"> |
| <Pause/> the same thing. |
| </Segment> |
| <Segment StartTime="1149.569" EndTime="1152.930" Participant="me006"> |
| <NonVocalSound Description="mike noise"/> |
| </Segment> |
| <Segment StartTime="1150.288" EndTime="1154.255" Participant="me018"> |
| So, so the upper part is <Emphasis> training </Emphasis> <Pause/> T_I-digits? |
| </Segment> |
| <Segment StartTime="1150.709" EndTime="1150.933" Participant="mn007"> |
| <VocalSound Description="inbreath"/> |
| </Segment> |
| <Segment StartTime="1150.933" EndTime="1151.204" Participant="mn007"> |
| So. |
| </Segment> |
| <Segment StartTime="1154.070" EndTime="1156.914" Participant="mn007"> |
| It's - it's the H_T_K results, I mean. So it's |
| </Segment> |
| <Segment StartTime="1156.914" EndTime="1158.974" Participant="mn007"> |
| <Pause/> H_T_K training testings |
| </Segment> |
| <Segment StartTime="1157.049" EndTime="1157.750" Participant="me018"> |
| <VocalSound Description="inbreath"/> Ah. |
| </Segment> |
| <Segment StartTime="1158.974" EndTime="1161.546" Participant="mn007"> |
| <Pause/> with different kind of features |
| </Segment> |
| <Segment StartTime="1161.546" EndTime="1163.425" Participant="mn007"> |
| <Pause/> and what appears in the |
| </Segment> |
| <Segment StartTime="1163.425" EndTime="1168.644" Participant="mn007"> |
| <Pause/> uh left column is <Pause/> the networks that are used for doing this. |
| </Segment> |
| <Segment StartTime="1172.989" EndTime="1174.330" Participant="me013"> |
| Hmm. |
| </Segment> |
| <Segment StartTime="1173.989" EndTime="1174.645" Participant="mn007"> |
| So. |
| </Segment> |
| <Segment StartTime="1174.645" EndTime="1175.903" Participant="mn007"> |
| Uh |
| </Segment> |
| <Segment StartTime="1177.570" EndTime="1178.917" Participant="mn007"> |
| Yeah. |
| </Segment> |
| <Segment StartTime="1178.917" EndTime="1180.304" Participant="me013"> |
| Well, |
| </Segment> |
| <Segment StartTime="1180.304" EndTime="1181.519" Participant="me013"> |
| What was is that i- |
| </Segment> |
| <Segment StartTime="1181.519" EndTime="1182.675" Participant="me013"> |
| What was it that you had |
| </Segment> |
| <Segment StartTime="1182.675" EndTime="1185.223" Participant="me013"> |
| <Pause/> done <Pause/> last week when you showed - Do you remember? |
| </Segment> |
| <Segment StartTime="1185.223" EndTime="1188.336" Participant="me013"> |
| Wh- when you showed me <Pause/> the - your table last week? |
| </Segment> |
| <Segment StartTime="1187.153" EndTime="1190.528" Participant="mn007"> |
| It- It was part of these results. |
| </Segment> |
| <Segment StartTime="1191.026" EndTime="1191.770" Participant="mn007"> |
| Mmm. <Comment Description="quietly"/> |
| </Segment> |
| <Segment StartTime="1193.429" EndTime="1194.124" Participant="mn007"> |
| Mmm. <Comment Description="quietly"/> |
| </Segment> |
| <Segment StartTime="1195.212" EndTime="1199.985" Participant="me018"> |
| So where is the baseline <Pause/> for the T_I-digits <Pause/> located in here? |
| </Segment> |
| <Segment StartTime="1200.832" EndTime="1203.456" Participant="mn007"> |
| You mean the H_T_K Aurora baseline? |
| </Segment> |
| <Segment StartTime="1203.456" EndTime="1204.317" Participant="me018"> |
| Yeah. |
| </Segment> |
| <Segment StartTime="1204.317" EndTime="1206.408" Participant="mn007"> |
| It's uh the one hundred number. |
| </Segment> |
| <Segment StartTime="1206.408" EndTime="1209.586" Participant="mn007"> |
| It's, well, all these numbers are the ratio |
| </Segment> |
| <Segment StartTime="1208.140" EndTime="1208.807" Participant="me018"> |
| Ah! |
| </Segment> |
| <Segment StartTime="1209.586" EndTime="1211.715" Participant="mn007"> |
| <Pause/> with respect to the baseline. |
| </Segment> |
| <Segment StartTime="1211.541" EndTime="1213.468" Participant="me018"> |
| Ah, O_K, O_K. |
| </Segment> |
| <Segment StartTime="1212.050" EndTime="1212.572" Participant="me013"> |
| <VocalSound Description="inbreath"/> |
| </Segment> |
| <Segment StartTime="1212.572" EndTime="1215.369" Participant="me013"> |
| So this is word - word error rate, so a <Emphasis> high </Emphasis> number is <Emphasis> bad. </Emphasis> |
| </Segment> |
| <Segment StartTime="1215.990" EndTime="1220.009" Participant="mn007"> |
| Yeah, this is <Pause/> a word error rate ratio. |
| </Segment> |
| <Segment StartTime="1216.304" EndTime="1216.975" Participant="fn002"> |
| Yeah. |
| </Segment> |
| <Segment StartTime="1220.009" EndTime="1220.778" Participant="mn007"> |
| Yeah. |
| </Segment> |
| <Segment StartTime="1220.470" EndTime="1221.846" Participant="me018"> |
| O_K, I see. |
| </Segment> |
| <Segment StartTime="1220.778" EndTime="1222.272" Participant="mn007"> |
| <Pause/> So, |
| </Segment> |
| <Segment StartTime="1222.272" EndTime="1224.042" Participant="mn007"> |
| seventy point two means that |
| </Segment> |
| <Segment StartTime="1224.042" EndTime="1227.135" Participant="mn007"> |
| <Pause/> we reduced the error rate uh by thirty - thirty percent. So. |
| </Segment> |
| <Segment StartTime="1226.705" EndTime="1228.644" Participant="me018"> |
| O_K, O_K, gotcha. |
| </Segment> |
| <Segment StartTime="1228.953" EndTime="1230.471" Participant="me013"> |
| <VocalSound Description="outbreath"/> O_K, |
| </Segment> |
| <Segment StartTime="1229.444" EndTime="1230.135" Participant="mn007"> |
| Hmm. |
| </Segment> |
| <Segment StartTime="1230.471" EndTime="1233.069" Participant="me013"> |
| <VocalSound Description="outbreath"/> so if we take uh |
| </Segment> |
| <Segment StartTime="1234.123" EndTime="1235.617" Participant="me013"> |
| um <Pause/> |
| </Segment> |
| <Segment StartTime="1235.617" EndTime="1236.470" Participant="me013"> |
| let's see |
| </Segment> |
| <Segment StartTime="1236.470" EndTime="1237.948" Participant="me013"> |
| <Pause/> P_L_P |
| </Segment> |
| <Segment StartTime="1237.948" EndTime="1243.975" Participant="me013"> |
| <Pause/> uh with on-line <Pause/> normalization and <Pause/> delta-del- so that's this thing you have circled here |
| </Segment> |
| <Segment StartTime="1243.975" EndTime="1245.999" Participant="me013"> |
| <Pause/> in the second column, |
| </Segment> |
| <Segment StartTime="1244.441" EndTime="1245.066" Participant="mn007"> |
| Yeah. |
| </Segment> |
| <Segment StartTime="1247.137" EndTime="1247.878" Participant="me013"> |
| um |
| </Segment> |
| <Segment StartTime="1247.878" EndTime="1250.892" Participant="me013"> |
| <Pause/> and "multi-English" refers to what? |
| </Segment> |
| <Segment StartTime="1250.892" EndTime="1252.354" Participant="mn007"> |
| To TIMIT. |
| </Segment> |
| <Segment StartTime="1253.970" EndTime="1254.827" Participant="mn007"> |
| Mmm. |
| </Segment> |
| <Segment StartTime="1254.827" EndTime="1260.754" Participant="mn007"> |
| Then you have <Pause/> uh M_F, <Pause/> M_S and M_E which are for French, Spanish and English. |
| </Segment> |
| <Segment StartTime="1260.754" EndTime="1263.302" Participant="mn007"> |
| <Pause/> And, <Pause/> yeah. |
| </Segment> |
| <Segment StartTime="1263.302" EndTime="1268.018" Participant="mn007"> |
| <Pause/> Actually I - <Pause/> I uh forgot to say that <Pause/> the multilingual net are trained |
| </Segment> |
| <Segment StartTime="1268.018" EndTime="1274.812" Participant="mn007"> |
| <Pause/> on <Pause/> uh <Pause/> features without the s- derivatives |
| </Segment> |
| <Segment StartTime="1274.812" EndTime="1279.306" Participant="mn007"> |
| <Pause/> uh but with <Pause/> increased frame numbers. Mmm. |
| </Segment> |
| <Segment StartTime="1281.410" EndTime="1282.484" Participant="me013"> |
| <VocalSound Description="inbreath"/> |
| </Segment> |
| <Segment StartTime="1284.320" EndTime="1287.913" Participant="mn007"> |
| And we can - we can see on the first line of the table that it - it - |
| </Segment> |
| <Segment StartTime="1287.913" EndTime="1292.123" Participant="mn007"> |
| <Pause/> it's <Pronounce> slightly </Pronounce> - slightly worse when we don't use delta but it's not - |
| </Segment> |
| <Segment StartTime="1292.123" EndTime="1294.252" Participant="mn007"> |
| <Pause/> not <Emphasis> that </Emphasis> much. |
| </Segment> |
| <Segment StartTime="1292.207" EndTime="1292.811" Participant="me013"> |
| Right. |
| </Segment> |
| <Segment StartTime="1292.811" EndTime="1294.076" Participant="me013"> |
| <VocalSound Description="inbreath"/> So w- w- |
| </Segment> |
| <Segment StartTime="1294.076" EndTime="1297.227" Participant="me013"> |
| So, I'm sorry. I missed that. What's M_F, M_S and M_E? |
| </Segment> |
| <Segment StartTime="1296.944" EndTime="1299.018" Participant="me018"> |
| Multi-French, Multi-Spanish |
| </Segment> |
| <Segment StartTime="1297.113" EndTime="1300.257" Participant="mn007"> |
| So. Multi-French, Multi-Spanish, and Multi-English. |
| </Segment> |
| <Segment StartTime="1299.251" EndTime="1303.219" Participant="me013"> |
| Uh O_K. So, it's <Pause/> uh <Pause/> <Emphasis> broader </Emphasis> vocabulary. |
| </Segment> |
| <Segment StartTime="1303.219" EndTime="1303.810" Participant="mn007"> |
| Yeah. |
| </Segment> |
| <Segment StartTime="1303.541" EndTime="1304.077" Participant="me013"> |
| Then - |
| </Segment> |
| <Segment StartTime="1304.538" EndTime="1305.358" Participant="me013"> |
| And - |
| </Segment> |
| <Segment StartTime="1305.358" EndTime="1307.328" Participant="me013"> |
| <Pause/> O_K so I <Emphasis> think </Emphasis> what I'm - |
| </Segment> |
| <Segment StartTime="1307.328" EndTime="1310.834" Participant="me013"> |
| what I saw in your smaller chart that I was thinking of was - |
| </Segment> |
| <Segment StartTime="1310.834" EndTime="1311.827" Participant="me013"> |
| was |
| </Segment> |
| <Segment StartTime="1311.827" EndTime="1316.766" Participant="me013"> |
| <Pause/> there were <Emphasis> some </Emphasis> numbers I saw, I think, that <Emphasis> included </Emphasis> these multiple languages |
| </Segment> |
| <Segment StartTime="1316.766" EndTime="1317.380" Participant="me013"> |
| <Pause/> and it - |
| </Segment> |
| <Segment StartTime="1317.380" EndTime="1318.735" Participant="me013"> |
| and I was seeing |
| </Segment> |
| <Segment StartTime="1318.735" EndTime="1320.141" Participant="me013"> |
| <Pause/> that it got <Emphasis> worse. </Emphasis> |
| </Segment> |
| <Segment StartTime="1320.141" EndTime="1324.062" Participant="me013"> |
| I - I think that was all it was. You had some very limited results that - at that point |
| </Segment> |
| <Segment StartTime="1323.172" EndTime="1324.140" Participant="mn007"> |
| Yeah. |
| </Segment> |
| <Segment StartTime="1324.062" EndTime="1329.700" Participant="me013"> |
| <Pause/> which showed <Pause/> having in these - these other languages. In fact it might have been just this last category, |
| </Segment> |
| <Segment StartTime="1327.068" EndTime="1329.358" Participant="me034"> |
| <VocalSound Description="sips coffee"/> |
| </Segment> |
| <Segment StartTime="1329.700" EndTime="1331.553" Participant="me013"> |
| <Pause/> having two languages broad |
| </Segment> |
| <Segment StartTime="1331.553" EndTime="1332.059" Participant="me013"> |
| that were - |
| </Segment> |
| <Segment StartTime="1332.059" EndTime="1333.721" Participant="me013"> |
| where - where English was <Emphasis> removed. </Emphasis> |
| </Segment> |
| <Segment StartTime="1333.721" EndTime="1335.637" Participant="me013"> |
| <Pause/> So that was cross language |
| </Segment> |
| <Segment StartTime="1335.637" EndTime="1338.181" Participant="me013"> |
| <Pause/> and the - and the result was quite <Emphasis> poor. </Emphasis> |
| </Segment> |
| <Segment StartTime="1338.747" EndTime="1339.290" Participant="me013"> |
| What I - |
| </Segment> |
| <Segment StartTime="1339.290" EndTime="1342.683" Participant="me013"> |
| <Pause/> we <Emphasis> hadn't </Emphasis> seen yet was that if you added <Emphasis> in </Emphasis> the English, it's <Emphasis> still </Emphasis> poor. |
| </Segment> |
| <Segment StartTime="1343.124" EndTime="1345.828" Participant="mn007"> |
| Yeah. <Pause/> Still poor. <VocalSound Description="laugh"/> |
| </Segment> |
| <Segment StartTime="1343.572" EndTime="1346.375" Participant="me013"> |
| <VocalSound Description="laugh"/> Uh <VocalSound Description="laugh"/> |
| </Segment> |
| <Segment StartTime="1346.375" EndTime="1346.862" Participant="me013"> |
| <VocalSound Description="inbreath"/> |
| </Segment> |
| <Segment StartTime="1346.862" EndTime="1349.501" Participant="me013"> |
| Um now, what's the noise condition |
| </Segment> |
| <Segment StartTime="1349.501" EndTime="1352.193" Participant="me013"> |
| <Pause/> um <Pause/> of the training data - |
| </Segment> |
| <Segment StartTime="1352.442" EndTime="1355.402" Participant="me013"> |
| Well, I think this is what you were explaining. The <Emphasis> noise </Emphasis> condition is the same - |
| </Segment> |
| <Segment StartTime="1355.402" EndTime="1357.863" Participant="me013"> |
| It's the same uh Aurora noises |
| </Segment> |
| <Segment StartTime="1357.503" EndTime="1358.216" Participant="me006"> |
| <NonVocalSound Description="mike noise"/> |
| </Segment> |
| <Segment StartTime="1357.767" EndTime="1358.362" Participant="mn007"> |
| Yeah. |
| </Segment> |
| <Segment StartTime="1358.438" EndTime="1359.847" Participant="me013"> |
| uh, in all these cases |
| </Segment> |
| <Segment StartTime="1359.844" EndTime="1360.481" Participant="mn007"> |
| Yeah. |
| </Segment> |
| <Segment StartTime="1359.847" EndTime="1360.534" Participant="me013"> |
| <Pause/> for the training. |
| </Segment> |
| <Segment StartTime="1361.167" EndTime="1364.285" Participant="me013"> |
| So there's not a <Pause/> statistical - sta- a strong st- |
| </Segment> |
| <Segment StartTime="1364.285" EndTime="1365.766" Participant="me013"> |
| <Pause/> statistically different |
| </Segment> |
| <Segment StartTime="1365.766" EndTime="1368.368" Participant="me013"> |
| <Pause/> noise characteristic between |
| </Segment> |
| <Segment StartTime="1368.234" EndTime="1370.832" Participant="mn007"> |
| No these are the s- s- s- same noises, yeah. |
| </Segment> |
| <Segment StartTime="1368.368" EndTime="1370.211" Participant="me013"> |
| <Pause/> uh the training and test |
| </Segment> |
| <Segment StartTime="1370.211" EndTime="1372.687" Participant="me013"> |
| <Pause/> and yet we're seeing <Emphasis> some </Emphasis> kind of effect - |
| </Segment> |
| <Segment StartTime="1370.832" EndTime="1373.073" Participant="mn007"> |
| <Pause/> At least - at least for the first - |
| </Segment> |
| <Segment StartTime="1373.073" EndTime="1375.562" Participant="mn007"> |
| <Pause/> for the well-matched, yeah. |
| </Segment> |
| <Segment StartTime="1374.224" EndTime="1375.808" Participant="me006"> |
| Well matched condition. |
| </Segment> |
| <Segment StartTime="1375.440" EndTime="1375.857" Participant="me013"> |
| Right. |
| </Segment> |
| <Segment StartTime="1375.857" EndTime="1381.621" Participant="me013"> |
| <Pause/> So there's <Emphasis> some </Emphasis> kind of a - a - an effect from having these - uh this broader coverage |
| </Segment> |
| <Segment StartTime="1381.621" EndTime="1383.095" Participant="me013"> |
| <Pause/> um |
| </Segment> |
| <Segment StartTime="1385.579" EndTime="1386.063" Participant="me013"> |
| <VocalSound Description="inbreath"/> |
| </Segment> |
| <Segment StartTime="1386.492" EndTime="1389.799" Participant="me013"> |
| Now I <Emphasis> guess </Emphasis> what we should try doing with this is try <Pause/> |
| </Segment> |
| <Segment StartTime="1389.799" EndTime="1392.835" Participant="me013"> |
| testing these on u- this same sort of thing on - |
| </Segment> |
| <Segment StartTime="1392.835" EndTime="1393.875" Participant="me013"> |
| <Pause/> you probably <Emphasis> must </Emphasis> have this |
| </Segment> |
| <Segment StartTime="1393.875" EndTime="1395.700" Participant="me013"> |
| <Pause/> <Emphasis> lined </Emphasis> up to <Emphasis> do. </Emphasis> To try the same t- |
| </Segment> |
| <Segment StartTime="1395.700" EndTime="1398.537" Participant="me013"> |
| <Pause/> with the <Emphasis> exact </Emphasis> same <Emphasis> training, </Emphasis> do testing on <Pause/> |
| </Segment> |
| <Segment StartTime="1398.537" EndTime="1399.856" Participant="me013"> |
| the other <Emphasis> languages. </Emphasis> |
| </Segment> |
| <Segment StartTime="1400.395" EndTime="1400.996" Participant="mn007"> |
| Mmm. |
| </Segment> |
| <Segment StartTime="1400.904" EndTime="1402.765" Participant="me013"> |
| On - on um - |
| </Segment> |
| <Segment StartTime="1404.174" EndTime="1405.429" Participant="me013"> |
| So. |
| </Segment> |
| <Segment StartTime="1406.332" EndTime="1410.109" Participant="me013"> |
| Um, oh I well, <Emphasis> wait </Emphasis> a minute. You <Emphasis> have </Emphasis> this <Emphasis> here, </Emphasis> for the <Emphasis> Italian. </Emphasis> <Emphasis> That's </Emphasis> right. <Pause/> O_K, so, |
| </Segment> |
| <Segment StartTime="1409.342" EndTime="1409.646" Participant="mn007"> |
| Yeah. |
| </Segment> |
| <Segment StartTime="1409.646" EndTime="1417.885" Participant="mn007"> |
| <Pause/> Yeah, so for the Italian the results are <VocalSound Description="outbreath"/> uh <Pause/> stranger um |
| </Segment> |
| <Segment StartTime="1410.109" EndTime="1411.247" Participant="me013"> |
| <Pause/> So. |
| </Segment> |
| <Segment StartTime="1413.732" EndTime="1414.523" Participant="fn002"> |
| <VocalSound Description="laugh"/> |
| </Segment> |
| <Segment StartTime="1417.885" EndTime="1420.342" Participant="mn007"> |
| <Pause/> Mmm. |
| </Segment> |
| <Segment StartTime="1420.342" EndTime="1423.631" Participant="mn007"> |
| <Pause/> So what appears is that perhaps Spanish is |
| </Segment> |
| <Segment StartTime="1423.631" EndTime="1425.583" Participant="mn007"> |
| <Pause/> not very close to Italian |
| </Segment> |
| <Segment StartTime="1425.447" EndTime="1426.916" Participant="me013"> |
| <VocalSound Description="laugh"/> |
| </Segment> |
| <Segment StartTime="1425.583" EndTime="1427.453" Participant="mn007"> |
| <Pause/> because uh, well, |
| </Segment> |
| <Segment StartTime="1427.453" EndTime="1430.272" Participant="mn007"> |
| <Pause/> when using the - the network trained only on Spanish it's - |
| </Segment> |
| <Segment StartTime="1430.272" EndTime="1432.278" Participant="mn007"> |
| <Pause/> the error rate is |
| </Segment> |
| <Segment StartTime="1432.278" EndTime="1435.133" Participant="mn007"> |
| <Pause/> almost uh twice |
| </Segment> |
| <Segment StartTime="1435.133" EndTime="1437.735" Participant="mn007"> |
| <Pause/> the baseline error rate. |
| </Segment> |
| <Segment StartTime="1437.637" EndTime="1438.239" Participant="me013"> |
| Mm-hmm. |
| </Segment> |
| <Segment StartTime="1438.250" EndTime="1442.233" Participant="mn007"> |
| Mmm. <VocalSound Description="inbreath"/> Uh. |
| </Segment> |
| <Segment StartTime="1440.933" EndTime="1441.360" Participant="me013"> |
| <VocalSound Description="inbreath"/> |
| </Segment> |
| <Segment StartTime="1441.360" EndTime="1442.559" Participant="me013"> |
| Well, I mean, let's see. |
| </Segment> |
| <Segment StartTime="1442.559" EndTime="1443.933" Participant="me013"> |
| <Pause/> Is there any difference in - |
| </Segment> |
| <Segment StartTime="1443.933" EndTime="1446.806" Participant="me013"> |
| <Pause/> So it's in <Pause/> the uh - |
| </Segment> |
| <Segment StartTime="1446.806" EndTime="1447.710" Participant="me013"> |
| <Pause/> So you're saying that |
| </Segment> |
| <Segment StartTime="1447.710" EndTime="1449.511" Participant="me013"> |
| <Pause/> when you <Emphasis> train </Emphasis> on English |
| </Segment> |
| <Segment StartTime="1449.511" EndTime="1454.137" Participant="me013"> |
| <Pause/> and <Pause/> uh <Pause/> and - and <Emphasis> test </Emphasis> on - |
| </Segment> |
| <Segment StartTime="1449.561" EndTime="1450.439" Participant="mn007"> |
| Yeah. |
| </Segment> |
| <Segment StartTime="1455.167" EndTime="1457.187" Participant="me013"> |
| No, you <Emphasis> don't </Emphasis> have training on English testing - <Comment Description="correcting himself"/> |
| </Segment> |
| <Segment StartTime="1455.877" EndTime="1459.881" Participant="mn007"> |
| There - there is - <Emphasis> another </Emphasis> difference, is that the noise - the noises are different. Well, |
| </Segment> |
| <Segment StartTime="1459.881" EndTime="1460.914" Participant="me013"> |
| In - in <Emphasis> what? </Emphasis> |
| </Segment> |
| <Segment StartTime="1461.076" EndTime="1463.125" Participant="mn007"> |
| For - for the Italian part I mean the |
| </Segment> |
| <Segment StartTime="1463.125" EndTime="1466.721" Participant="mn007"> |
| <Pause/> uh <Pause/> the um |
| </Segment> |
| <Segment StartTime="1466.721" EndTime="1473.128" Participant="mn007"> |
| <Pause/> networks are trained with noise from <Pause/> Aurora - T_I-<Emphasis> digits, </Emphasis> mmm. |
| </Segment> |
| <Segment StartTime="1471.238" EndTime="1472.482" Participant="fn002"> |
| Aurora-two. <Comment Description="completing [mn007]'s point"/> |
| </Segment> |
| <Segment StartTime="1473.128" EndTime="1475.170" Participant="mn007"> |
| <Pause/> <VocalSound Description="inbreath"/> Yeah. |
| </Segment> |
| <Segment StartTime="1474.849" EndTime="1476.940" Participant="me013"> |
| And the noise is different <Uncertain> in th- </Uncertain> |
| </Segment> |
| <Segment StartTime="1475.170" EndTime="1476.850" Participant="mn007"> |
| <Pause/> And perhaps the noise are |
| </Segment> |
| <Segment StartTime="1476.850" EndTime="1480.870" Participant="mn007"> |
| <Pause/> quite different from the noises <Pause/> in the speech that <Uncertain> Italian </Uncertain>. |
| </Segment> |
| <Segment StartTime="1480.657" EndTime="1481.271" Participant="me013"> |
| <VocalSound Description="inbreath"/> |
| </Segment> |
| <Segment StartTime="1480.870" EndTime="1482.506" Participant="mn007"> |
| <Pause/> And - |
| </Segment> |
| <Segment StartTime="1481.271" EndTime="1482.898" Participant="me013"> |
| Do we have any um |
| </Segment> |
| <Segment StartTime="1482.898" EndTime="1484.486" Participant="me013"> |
| <Pause/> test sets |
| </Segment> |
| <Segment StartTime="1484.486" EndTime="1489.582" Participant="me013"> |
| <Pause/> uh in <Pause/> any other language that um |
| </Segment> |
| <Segment StartTime="1491.371" EndTime="1494.840" Participant="me013"> |
| have the <Emphasis> same </Emphasis> noise as in <Pause/> the Aurora? |
| </Segment> |
| <Segment StartTime="1495.739" EndTime="1496.496" Participant="fn002"> |
| Mmm, no. |
| </Segment> |
| <Segment StartTime="1497.472" EndTime="1498.110" Participant="mn007"> |
| No. |
| </Segment> |
| <Segment StartTime="1500.974" EndTime="1502.149" Participant="me018"> |
| Can I ask something real quick? |
| </Segment> |
| <Segment StartTime="1502.149" EndTime="1504.155" Participant="me018"> |
| <Pause/> In - in the <Emphasis> upper </Emphasis> part - |
| </Segment> |
| <Segment StartTime="1504.155" EndTime="1506.300" Participant="me018"> |
| <Pause/> in the <Emphasis> English </Emphasis> <Pause/> stuff, |
| </Segment> |
| <Segment StartTime="1506.300" EndTime="1509.661" Participant="me018"> |
| <Pause/> it <Emphasis> looks </Emphasis> like the <Emphasis> very </Emphasis> <Emphasis> best </Emphasis> number is sixty point nine? |
| </Segment> |
| <Segment StartTime="1509.661" EndTime="1511.504" Participant="me018"> |
| <Pause/> and that's in the uh - |
| </Segment> |
| <Segment StartTime="1511.504" EndTime="1515.660" Participant="me018"> |
| <Pause/> the third <Pause/> section in the upper part under P_L_P J_RASTA, |
| </Segment> |
| <Segment StartTime="1515.660" EndTime="1516.907" Participant="me018"> |
| <Pause/> sort of the middle column? |
| </Segment> |
| <Segment StartTime="1518.168" EndTime="1518.715" Participant="mn007"> |
| Yeah. |
| </Segment> |
| <Segment StartTime="1519.018" EndTime="1521.356" Participant="me018"> |
| I- is <Emphasis> that </Emphasis> <Pause/> a <Emphasis> noisy </Emphasis> condition? |
| </Segment> |
| <Segment StartTime="1521.610" EndTime="1522.217" Participant="mn007"> |
| Yeah. |
| </Segment> |
| <Segment StartTime="1523.864" EndTime="1526.602" Participant="me018"> |
| So that's <Emphasis> matched </Emphasis> training? Is <Emphasis> that </Emphasis> what that <Emphasis> is? </Emphasis> |
| </Segment> |
| <Segment StartTime="1526.864" EndTime="1530.954" Participant="mn007"> |
| It's - no, the third part, so it's uh <Pause/> highly mismatched. |
| </Segment> |
| <Segment StartTime="1530.954" EndTime="1534.781" Participant="mn007"> |
| <Pause/> So. Training and <Pause/> test noise are different. |
| </Segment> |
| <Segment StartTime="1532.288" EndTime="1534.813" Participant="me018"> |
| So - why do you get your <Emphasis> best </Emphasis> <Emphasis> number </Emphasis> |
| </Segment> |
| <Segment StartTime="1535.811" EndTime="1536.630" Participant="me018"> |
| in - |
| </Segment> |
| <Segment StartTime="1537.324" EndTime="1539.778" Participant="me018"> |
| Wouldn't you get your <Emphasis> best </Emphasis> number in the <Emphasis> clean </Emphasis> case? |
| </Segment> |
| <Segment StartTime="1539.750" EndTime="1541.105" Participant="me034"> |
| Well, it's relative to the |
| </Segment> |
| <Segment StartTime="1541.762" EndTime="1542.134" Participant="me034"> |
| um |
| </Segment> |
| <Segment StartTime="1542.134" EndTime="1543.385" Participant="me034"> |
| <Pause/> baseline mismatching |
| </Segment> |
| <Segment StartTime="1542.307" EndTime="1542.731" Participant="mn007"> |
| Yeah. |
| </Segment> |
| <Segment StartTime="1543.385" EndTime="1545.369" Participant="me034"> |
| <VocalSound Description="laugh"/> |
| </Segment> |
| <Segment StartTime="1543.773" EndTime="1545.382" Participant="me018"> |
| Ah, O_K so these are not - |
| </Segment> |
| <Segment StartTime="1543.805" EndTime="1544.290" Participant="mn007"> |
| Yeah. |
| </Segment> |
| <Segment StartTime="1545.382" EndTime="1547.529" Participant="me018"> |
| <Pause/> O_K, <Pause/> alright, I see. |
| </Segment> |
| <Segment StartTime="1546.219" EndTime="1546.805" Participant="me034"> |
| Yeah. |
| </Segment> |
| <Segment StartTime="1548.259" EndTime="1549.311" Participant="mn007"> |
| Yeah. <VocalSound Description="breath-laugh"/> |
| </Segment> |
| <Segment StartTime="1548.818" EndTime="1549.313" Participant="me018"> |
| O_K. |
| </Segment> |
| <Segment StartTime="1550.783" EndTime="1552.739" Participant="me018"> |
| And then - so, in the - in the um - |
| </Segment> |
| <Segment StartTime="1552.739" EndTime="1556.190" Participant="me018"> |
| <Pause/> in the <Pause/> <Emphasis> non-mismatched </Emphasis> <Emphasis> clean </Emphasis> case, |
| </Segment> |
| <Segment StartTime="1556.190" EndTime="1558.973" Participant="me018"> |
| <Pause/> your best one was under M_F_C_C? |
| </Segment> |
| <Segment StartTime="1558.973" EndTime="1561.882" Participant="me018"> |
| <Pause/> That sixty-one point four? |
| </Segment> |
| <Segment StartTime="1561.882" EndTime="1566.113" Participant="mn007"> |
| Yeah. <Pause/> But it's not a <Emphasis> clean </Emphasis> case. It's <Pause/> a noisy case but |
| </Segment> |
| <Segment StartTime="1566.113" EndTime="1569.362" Participant="mn007"> |
| <Pause/> uh training and test noises are the same. |
| </Segment> |
| <Segment StartTime="1570.549" EndTime="1572.193" Participant="me018"> |
| Oh! So this upper third? |
| </Segment> |
| <Segment StartTime="1571.075" EndTime="1572.674" Participant="mn007"> |
| So - Yeah. |
| </Segment> |
| <Segment StartTime="1573.024" EndTime="1574.849" Participant="me018"> |
| Uh that's still noisy? |
| </Segment> |
| <Segment StartTime="1574.936" EndTime="1575.566" Participant="mn007"> |
| Yeah. |
| </Segment> |
| <Segment StartTime="1575.466" EndTime="1576.422" Participant="me018"> |
| Ah, O_K. |
| </Segment> |
| <Segment StartTime="1578.060" EndTime="1584.384" Participant="mn007"> |
| So it's always noisy basically, <Pause/> and, <Pause/> well, the - |
| </Segment> |
| <Segment StartTime="1580.030" EndTime="1580.718" Participant="me018"> |
| Mm-hmm. |
| </Segment> |
| <Segment StartTime="1583.148" EndTime="1584.016" Participant="me018"> |
| I see. |
| </Segment> |
| <Segment StartTime="1584.872" EndTime="1585.770" Participant="mn007"> |
| Mmm. |
| </Segment> |
| <Segment StartTime="1589.595" EndTime="1589.866" Participant="me013"> |
| <VocalSound Description="inbreath"/> |
| </Segment> |
| <Segment StartTime="1589.866" EndTime="1590.281" Participant="me013"> |
| O_K? |
| </Segment> |
| <Segment StartTime="1590.054" EndTime="1595.155" Participant="me006"> |
| <NonVocalSound Description="mike noise"/> |
| </Segment> |
| <Segment StartTime="1591.673" EndTime="1593.552" Participant="me013"> |
| Um <Pause/> |
| </Segment> |
| <Segment StartTime="1593.552" EndTime="1598.522" Participant="me013"> |
| So uh, I think this will take some <Pause/> looking at, thinking about. But, |
| </Segment> |
| <Segment StartTime="1598.522" EndTime="1600.017" Participant="me013"> |
| <Pause/> what is uh - |
| </Segment> |
| <Segment StartTime="1599.690" EndTime="1601.649" Participant="me006"> |
| <NonVocalSound Description="mike noise"/> |
| </Segment> |
| <Segment StartTime="1600.017" EndTime="1602.185" Participant="me013"> |
| what is <Emphasis> currently </Emphasis> running, that's - |
| </Segment> |
| <Segment StartTime="1602.185" EndTime="1603.125" Participant="me013"> |
| uh, i- that - |
| </Segment> |
| <Segment StartTime="1603.125" EndTime="1605.330" Participant="me013"> |
| just filling in the holes here or - or - ? <Comment Description="leading question"/> |
| </Segment> |
| <Segment StartTime="1604.897" EndTime="1608.330" Participant="mn007"> |
| Uh, no we don't plan to fill the holes but |
| </Segment> |
| <Segment StartTime="1605.330" EndTime="1606.627" Participant="me013"> |
| <Pause/> pretty much? |
| </Segment> |
| <Segment StartTime="1607.801" EndTime="1608.545" Participant="me013"> |
| O_K. |
| </Segment> |
| <Segment StartTime="1608.330" EndTime="1610.499" Participant="mn007"> |
| <Pause/> actually there is something important, |
| </Segment> |
| <Segment StartTime="1610.499" EndTime="1616.697" Participant="mn007"> |
| <Pause/> is that <Pause/> um we made a lot of assumption concerning the on-line normalization |
| </Segment> |
| <Segment StartTime="1616.697" EndTime="1619.642" Participant="mn007"> |
| <Pause/> and we just noticed |
| </Segment> |
| <Segment StartTime="1619.642" EndTime="1621.757" Participant="mn007"> |
| <Pause/> uh recently that |
| </Segment> |
| <Segment StartTime="1621.757" EndTime="1626.265" Participant="mn007"> |
| <Pause/> uh the <Pause/> approach that we were using |
| </Segment> |
| <Segment StartTime="1626.265" EndTime="1628.235" Participant="mn007"> |
| <Pause/> was not <Pause/> uh |
| </Segment> |
| <Segment StartTime="1628.235" EndTime="1630.692" Participant="mn007"> |
| <Pause/> leading to very good results |
| </Segment> |
| <Segment StartTime="1630.692" EndTime="1633.547" Participant="mn007"> |
| <Pause/> when we <Pause/> used the straight features to H_T_K. |
| </Segment> |
| <Segment StartTime="1635.612" EndTime="1637.855" Participant="mn007"> |
| Um <Pause/> |
| </Segment> |
| <Segment StartTime="1637.438" EndTime="1638.409" Participant="me006"> |
| <NonVocalSound Description="mike noise"/> |
| </Segment> |
| <Segment StartTime="1637.855" EndTime="1640.468" Participant="mn007"> |
| <Pause/> Mmm. So basically d- |
| </Segment> |
| <Segment StartTime="1640.468" EndTime="1643.124" Participant="mn007"> |
| <Pause/> if you look at the - at the left of the table, |
| </Segment> |
| <Segment StartTime="1643.124" EndTime="1646.341" Participant="mn007"> |
| <Pause/> the first uh row, |
| </Segment> |
| <Segment StartTime="1646.341" EndTime="1649.756" Participant="mn007"> |
| <Pause/> with eighty-six, one hundred, and forty-three and seventy-five, |
| </Segment> |
| <Segment StartTime="1651.225" EndTime="1653.999" Participant="mn007"> |
| these are the results we obtained for Italian |
| </Segment> |
| <Segment StartTime="1653.999" EndTime="1660.259" Participant="mn007"> |
| <Pause/> uh with <Pause/> straight <Pause/> mmm, P_L_P features |
| </Segment> |
| <Segment StartTime="1654.603" EndTime="1655.345" Participant="me006"> |
| <NonVocalSound Description="mike noise"/> |
| </Segment> |
| <Segment StartTime="1660.259" EndTime="1663.183" Participant="mn007"> |
| <Pause/> using on-line normalization. |
| </Segment> |
| <Segment StartTime="1661.850" EndTime="1665.980" Participant="me006"> |
| <NonVocalSound Description="mike noise"/> |
| </Segment> |
| <Segment StartTime="1664.083" EndTime="1664.670" Participant="me013"> |
| Mm-hmm. |
| </Segment> |
| <Segment StartTime="1665.419" EndTime="1666.647" Participant="mn007"> |
| Mmm. |
| </Segment> |
| <Segment StartTime="1668.577" EndTime="1676.564" Participant="mn007"> |
| And the, mmm - what's <Pause/> in the table, just <Pause/> at the left of the P_L_P twelve <Pause/> on-line normalization column, |
| </Segment> |
| <Segment StartTime="1676.564" EndTime="1680.975" Participant="mn007"> |
| <Pause/> so, the numbers seventy-nine, fifty-four and <Pause/> uh forty-two |
| </Segment> |
| <Segment StartTime="1680.474" EndTime="1685.818" Participant="me006"> |
| <NonVocalSound Description="mike noise"/> |
| </Segment> |
| <Segment StartTime="1680.975" EndTime="1688.347" Participant="mn007"> |
| <Pause/> are the results obtained by uh Pratibha with <Pause/> uh his on-line normalization - uh <Emphasis> her </Emphasis> on-line normalization approach. |
| </Segment> |
| <Segment StartTime="1687.767" EndTime="1689.944" Participant="me018"> |
| Where is that? seventy-nine, fifty |
| </Segment> |
| <Segment StartTime="1689.782" EndTime="1691.140" Participant="fn002"> |
| Fifty-one? This - <Comment Description="whispered"/> |
| </Segment> |
| <Segment StartTime="1689.944" EndTime="1693.891" Participant="me013"> |
| Uh, it's just sort of sitting right on the uh - the column line. |
| </Segment> |
| <Segment StartTime="1690.640" EndTime="1691.181" Participant="mn007"> |
| So. |
| </Segment> |
| <Segment StartTime="1694.433" EndTime="1695.644" Participant="me013"> |
| Uh. <Pause/> |
| </Segment> |
| <Segment StartTime="1694.803" EndTime="1696.256" Participant="me018"> |
| Oh I see, O_K. |
| </Segment> |
| <Segment StartTime="1695.053" EndTime="1695.943" Participant="mn007"> |
| Just - uh |
| </Segment> |
| <Segment StartTime="1695.644" EndTime="1696.256" Participant="me013"> |
| Yeah. |
| </Segment> |
| <Segment StartTime="1696.256" EndTime="1696.750" Participant="mn007"> |
| Yeah. |
| </Segment> |
| <Segment StartTime="1696.750" EndTime="1703.526" Participant="mn007"> |
| <Pause/> So these are the results of <Pause/> O_G_I with <Pause/> on-line normalization and straight features to H_T_K. |
| </Segment> |
| <Segment StartTime="1703.526" EndTime="1706.251" Participant="mn007"> |
| <Pause/> And the previous result, eighty-six and so on, |
| </Segment> |
| <Segment StartTime="1705.991" EndTime="1706.594" Participant="me013"> |
| Yes. |
| </Segment> |
| <Segment StartTime="1706.251" EndTime="1710.118" Participant="mn007"> |
| <Pause/> are with our <Pause/> features straight to H_T_K. So |
| </Segment> |
| <Segment StartTime="1708.511" EndTime="1709.141" Participant="me013"> |
| Yes. |
| </Segment> |
| <Segment StartTime="1710.118" EndTime="1712.865" Participant="mn007"> |
| <Pause/> what we see that - is - there is that um |
| </Segment> |
| <Segment StartTime="1712.865" EndTime="1717.744" Participant="mn007"> |
| <Pause/> uh the way we were doing this was not correct, but <Pause/> still |
| </Segment> |
| <Segment StartTime="1717.744" EndTime="1720.270" Participant="mn007"> |
| <Pause/> the networks <Pause/> are very good. |
| </Segment> |
| <Segment StartTime="1718.615" EndTime="1719.585" Participant="me006"> |
| <NonVocalSound Description="mike noise"/> |
| </Segment> |
| <Segment StartTime="1720.270" EndTime="1722.131" Participant="mn007"> |
| <Pause/> When we use the networks |
| </Segment> |
| <Segment StartTime="1722.131" EndTime="1724.534" Participant="mn007"> |
| <Pause/> our number are better that |
| </Segment> |
| <Segment StartTime="1723.027" EndTime="1724.389" Participant="fn002"> |
| We improve. <Comment Description="completing [mn007]'s point"/> |
| </Segment> |
| <Segment StartTime="1724.534" EndTime="1726.988" Participant="mn007"> |
| <Pause/> uh Pratibha results. |
| </Segment> |
| <Segment StartTime="1725.294" EndTime="1728.013" Participant="me013"> |
| So, do you know what was <Emphasis> wrong </Emphasis> with the on-line normalization, or - ? <Comment Description="leading question"/> |
| </Segment> |
| <Segment StartTime="1728.145" EndTime="1730.971" Participant="mn007"> |
| Yeah. There were diff- there were different things and |
| </Segment> |
| <Segment StartTime="1730.971" EndTime="1734.730" Participant="mn007"> |
| <Pause/> basically, <Pause/> the first thing is the mmm, |
| </Segment> |
| <Segment StartTime="1732.826" EndTime="1733.911" Participant="me006"> |
| <NonVocalSound Description="mike noise"/> |
| </Segment> |
| <Segment StartTime="1734.730" EndTime="1743.552" Participant="mn007"> |
| <Pause/> alpha uh <Pause/> value. So, the recursion <Pause/> uh <Pause/> part. |
| </Segment> |
| <Segment StartTime="1743.552" EndTime="1745.413" Participant="mn007"> |
| <Pause/> um, |
| </Segment> |
| <Segment StartTime="1745.413" EndTime="1750.943" Participant="mn007"> |
| <Pause/> I used point five percent, <Pause/> which was the default value in the - |
| </Segment> |
| <Segment StartTime="1750.943" EndTime="1752.713" Participant="mn007"> |
| <Pause/> in the programs here. |
| </Segment> |
| <Segment StartTime="1752.713" EndTime="1755.731" Participant="mn007"> |
| <Pause/> And Pratibha used <Emphasis> five </Emphasis> percent. |
| </Segment> |
| <Segment StartTime="1755.731" EndTime="1758.170" Participant="mn007"> |
| So it adapts more <Pause/> quickly |
| </Segment> |
| <Segment StartTime="1755.731" EndTime="1757.788" Participant="me013"> |
| Uh-huh. <VocalSound Description="laugh"/> |
| </Segment> |
| <Segment StartTime="1757.788" EndTime="1758.948" Participant="me013"> |
| Yes. Yeah. |
| </Segment> |
| <Segment StartTime="1758.170" EndTime="1763.212" Participant="mn007"> |
| <Pause/> Um, but, yeah. I assume that this was not important because |
| </Segment> |
| <Segment StartTime="1763.212" EndTime="1767.892" Participant="mn007"> |
| <Pause/> uh previous results from - from Dan and - show that basically |
| </Segment> |
| <Segment StartTime="1767.892" EndTime="1776.421" Participant="mn007"> |
| <Pause/> the <Pause/> both - both values g- give the same - same <Pause/> uh results. |
| </Segment> |
| <Segment StartTime="1776.421" EndTime="1781.514" Participant="mn007"> |
| <Pause/> It was true on uh <Pause/> T_I-digits but it's not true on Italian. |
| </Segment> |
| <Segment StartTime="1780.123" EndTime="1780.954" Participant="me013"> |
| Mm-hmm. |
| </Segment> |
| <Segment StartTime="1781.514" EndTime="1787.534" Participant="mn007"> |
| <Pause/> Uh, second thing is the initialization of the <Pause/> stuff. Actually, |
| </Segment> |
| <Segment StartTime="1787.534" EndTime="1794.481" Participant="mn007"> |
| <Pause/> uh what we were doing is to start the recursion from the beginning of the <Pause/> utterance. |
| </Segment> |
| <Segment StartTime="1794.481" EndTime="1802.028" Participant="mn007"> |
| <Pause/> And using initial values that are the global mean and variances <Pause/> measured across the <Uncertain> whole </Uncertain> database. |
| </Segment> |
| <Segment StartTime="1799.886" EndTime="1802.341" Participant="me013"> |
| Right. <Pause/> Right. |
| </Segment> |
| <Segment StartTime="1802.028" EndTime="1808.967" Participant="mn007"> |
| And Pratibha did something different is that he - uh she initialed the um values of the mean and variance |
| </Segment> |
| <Segment StartTime="1808.967" EndTime="1816.083" Participant="mn007"> |
| <Pause/> by computing <Pause/> this on the <Pause/> twenty-five first frames of each utterance. |
| </Segment> |
| <Segment StartTime="1817.786" EndTime="1823.438" Participant="mn007"> |
| Mmm. There were other minor differences, the fact that |
| </Segment> |
| <Segment StartTime="1823.438" EndTime="1827.288" Participant="mn007"> |
| <Pause/> she used fifteen <Uncertain> dissities </Uncertain> instead s- instead of thirteen, |
| </Segment> |
| <Segment StartTime="1827.288" EndTime="1831.281" Participant="mn007"> |
| <Pause/> and that she used C_zero instead of log energy. |
| </Segment> |
| <Segment StartTime="1831.710" EndTime="1835.205" Participant="mn007"> |
| Uh, but the main differences concerns the recursion. |
| </Segment> |
| <Segment StartTime="1835.205" EndTime="1836.506" Participant="mn007"> |
| <Pause/> So. |
| </Segment> |
| <Segment StartTime="1836.506" EndTime="1838.457" Participant="mn007"> |
| <Pause/> Uh, I changed the code |
| </Segment> |
| <Segment StartTime="1838.457" EndTime="1843.842" Participant="mn007"> |
| <Pause/> uh and now we have a baseline that's similar to the O_G_I baseline. |
| </Segment> |
| <Segment StartTime="1843.842" EndTime="1844.601" Participant="me013"> |
| O_K. |
| </Segment> |
| <Segment StartTime="1844.956" EndTime="1847.933" Participant="mn007"> |
| We - It - it's <Pronounce> slightly </Pronounce> <Pause/> uh different because |
| </Segment> |
| <Segment StartTime="1847.933" EndTime="1851.402" Participant="mn007"> |
| <Pause/> I don't exactly initialize the same way she does. |
| </Segment> |
| <Segment StartTime="1851.402" EndTime="1853.246" Participant="mn007"> |
| <Pause/> Actually I start, |
| </Segment> |
| <Segment StartTime="1853.246" EndTime="1857.818" Participant="mn007"> |
| <Pause/> mmm, I don't wait to a fifteen - twenty-five - twenty-five frames |
| </Segment> |
| <Segment StartTime="1857.818" EndTime="1860.122" Participant="mn007"> |
| <Pause/> before computing a mean and the variance |
| </Segment> |
| <Segment StartTime="1860.122" EndTime="1863.519" Participant="mn007"> |
| <Pause/> to e- to - to start the recursion. |
| </Segment> |
| <Segment StartTime="1860.158" EndTime="1860.854" Participant="me034"> |
| Mm-hmm. |
| </Segment> |
| <Segment StartTime="1862.455" EndTime="1863.136" Participant="me013"> |
| Yeah. |
| </Segment> |
| <Segment StartTime="1863.519" EndTime="1866.175" Participant="mn007"> |
| <Pause/> I - I use the on-line scheme |
| </Segment> |
| <Segment StartTime="1866.175" EndTime="1870.512" Participant="mn007"> |
| <Pause/> and only start the re- recursion after the twenty-five - <Pause/> twenty-fifth frame. |
| </Segment> |
| <Segment StartTime="1870.512" EndTime="1872.104" Participant="mn007"> |
| But, well it's similar. |
| </Segment> |
| <Segment StartTime="1872.104" EndTime="1873.893" Participant="mn007"> |
| <Pause/> So |
| </Segment> |
| <Segment StartTime="1873.893" EndTime="1884.585" Participant="mn007"> |
| <Pause/> uh I retrained <Pause/> the networks with <Pause/> these - well, the - the - the networks are retaining with these new <Pause/> features. |
| </Segment> |
| <Segment StartTime="1881.834" EndTime="1882.435" Participant="me013"> |
| Mm-hmm. |
| </Segment> |
| <Segment StartTime="1884.585" EndTime="1886.988" Participant="mn007"> |
| <Pause/> And, yeah. |
| </Segment> |
| <Segment StartTime="1889.138" EndTime="1889.745" Participant="me013"> |
| O_K. |
| </Segment> |
| <Segment StartTime="1891.036" EndTime="1892.681" Participant="mn007"> |
| So basically what I expect is that |
| </Segment> |
| <Segment StartTime="1892.681" EndTime="1895.548" Participant="mn007"> |
| <Pause/> these numbers will a little bit go down but |
| </Segment> |
| <Segment StartTime="1895.548" EndTime="1897.408" Participant="mn007"> |
| <Pause/> perhaps not - not so much |
| </Segment> |
| <Segment StartTime="1897.408" EndTime="1901.386" Participant="mn007"> |
| <Pause/> because <Pause/> I think the neural networks learn perhaps |
| </Segment> |
| <Segment StartTime="1897.559" EndTime="1898.741" Participant="me013"> |
| Right. |
| </Segment> |
| <Segment StartTime="1901.386" EndTime="1902.240" Participant="mn007"> |
| <Pause/> to - |
| </Segment> |
| <Segment StartTime="1902.240" EndTime="1907.946" Participant="mn007"> |
| <Pause/> even if the features are not <Pause/> normalized. It - it will learn how to normalize and - <VocalSound Description="inbreath"/> |
| </Segment> |
| <Segment StartTime="1902.903" EndTime="1903.564" Participant="me013"> |
| Right. |
| </Segment> |
| <Segment StartTime="1908.162" EndTime="1909.409" Participant="me013"> |
| O_K, but I think that |
| </Segment> |
| <Segment StartTime="1908.763" EndTime="1909.568" Participant="mn007"> |
| Mmm. |
| </Segment> |
| <Segment StartTime="1909.409" EndTime="1911.610" Participant="me013"> |
| <Pause/> given the pressure of time we probably want to draw - |
| </Segment> |
| <Segment StartTime="1911.610" EndTime="1915.802" Participant="me013"> |
| because of <Emphasis> that </Emphasis> <Pause/> <Emphasis> especially, </Emphasis> we wanna draw some <Emphasis> conclusions </Emphasis> from this, do some <Emphasis> reductions </Emphasis> |
| </Segment> |
| <Segment StartTime="1915.802" EndTime="1917.555" Participant="me013"> |
| <Pause/> in what we're <Emphasis> looking </Emphasis> at, |
| </Segment> |
| <Segment StartTime="1916.365" EndTime="1917.171" Participant="mn007"> |
| Yeah. |
| </Segment> |
| <Segment StartTime="1917.555" EndTime="1922.289" Participant="me013"> |
| <Pause/> and make some <Emphasis> strong </Emphasis> decisions for what we're gonna do <Emphasis> testing </Emphasis> on before next <Emphasis> week. </Emphasis> |
| </Segment> |
| <Segment StartTime="1922.289" EndTime="1923.930" Participant="me013"> |
| So do you - are you - w- |
| </Segment> |
| <Segment StartTime="1922.289" EndTime="1923.781" Participant="mn007"> |
| Yeah <VocalSound Description="inbreath"/> <Uncertain> I'd </Uncertain> - |
| </Segment> |
| <Segment StartTime="1923.930" EndTime="1928.741" Participant="me013"> |
| did you have something going on, on the side, with uh multi-band <Pause/> or - on - on this, or - ? <Comment Description="leading question"/> |
| </Segment> |
| <Segment StartTime="1927.896" EndTime="1933.371" Participant="mn007"> |
| No, I - we plan to start this uh so, act- actually we have discussed uh |
| </Segment> |
| <Segment StartTime="1933.371" EndTime="1935.951" Participant="mn007"> |
| <Pause/> <Uncertain NumSyllables="1"> @@ </Uncertain> um, these - |
| </Segment> |
| <Segment StartTime="1935.951" EndTime="1937.473" Participant="mn007"> |
| <Pause/> what we could do |
| </Segment> |
| <Segment StartTime="1937.473" EndTime="1940.220" Participant="mn007"> |
| <Pause/> more as a - as a research and - |
| </Segment> |
| <Segment StartTime="1940.220" EndTime="1943.292" Participant="mn007"> |
| <Pause/> and <Pause/> we were thinking perhaps that |
| </Segment> |
| <Segment StartTime="1943.292" EndTime="1946.870" Participant="mn007"> |
| <Pause/> uh <Pause/> the way we use the tandem is not - |
| </Segment> |
| <Segment StartTime="1946.870" EndTime="1951.297" Participant="mn007"> |
| <Pause/> Uh, well, there is basically perhaps a flaw in the - in the - the stuff because |
| </Segment> |
| <Segment StartTime="1951.297" EndTime="1954.080" Participant="mn007"> |
| <Pause/> we <Pause/> trained the networks - |
| </Segment> |
| <Segment StartTime="1954.080" EndTime="1956.068" Participant="mn007"> |
| If we trained the networks on the - on |
| </Segment> |
| <Segment StartTime="1956.068" EndTime="1960.088" Participant="mn007"> |
| <Pause/> a language and a t- or a specific <Pause/> task, |
| </Segment> |
| <Segment StartTime="1957.069" EndTime="1958.163" Participant="me006"> |
| <NonVocalSound Description="mike noise"/> |
| </Segment> |
| <Segment StartTime="1959.983" EndTime="1960.824" Participant="me013"> |
| Mm-hmm. |
| </Segment> |
| <Segment StartTime="1960.088" EndTime="1967.333" Participant="mn007"> |
| <Pause/> um, what we ask is - to the network - is to put the bound- the decision boundaries somewhere in the space. |
| </Segment> |
| <Segment StartTime="1967.333" EndTime="1974.189" Participant="mn007"> |
| <Pause/> And uh <Pause/> mmm and ask the network to put one, |
| </Segment> |
| <Segment StartTime="1967.731" EndTime="1968.504" Participant="me013"> |
| Mmm. |
| </Segment> |
| <Segment StartTime="1974.189" EndTime="1982.025" Participant="mn007"> |
| <Pause/> at one side of the - for - for a particular phoneme at one side of the boundary - decision boundary and one for another phoneme at the other side. |
| </Segment> |
| <Segment StartTime="1982.025" EndTime="1988.928" Participant="mn007"> |
| <Pause/> And <Pause/> so there is kind of reduction of the information there that's not correct because if we change task |
| </Segment> |
| <Segment StartTime="1988.928" EndTime="1992.460" Participant="mn007"> |
| <Pause/> and if the phonemes are not in the same context in the new task, |
| </Segment> |
| <Segment StartTime="1992.460" EndTime="1996.399" Participant="mn007"> |
| <Pause/> obviously the <Pause/> decision boundaries are not - |
| </Segment> |
| <Segment StartTime="1996.399" EndTime="2000.176" Participant="mn007"> |
| <Pause/> should not be at the same <Pause/> place. |
| </Segment> |
| <Segment StartTime="2002.964" EndTime="2004.619" Participant="mn007"> |
| But the way the feature gives - |
| </Segment> |
| <Segment StartTime="2002.993" EndTime="2003.411" Participant="me013"> |
| I di- |
| </Segment> |
| <Segment StartTime="2004.619" EndTime="2008.305" Participant="mn007"> |
| The - the way the network gives the features is that it reduce completely the - |
| </Segment> |
| <Segment StartTime="2008.305" EndTime="2010.474" Participant="mn007"> |
| <Pause/> it removes completely the information - |
| </Segment> |
| <Segment StartTime="2010.474" EndTime="2013.293" Participant="mn007"> |
| <Pause/> a lot of information from the - the features |
| </Segment> |
| <Segment StartTime="2013.293" EndTime="2015.403" Participant="mn007"> |
| <Pause/> by uh |
| </Segment> |
| <Segment StartTime="2015.403" EndTime="2017.987" Participant="mn007"> |
| <Pause/> uh |
| </Segment> |
| <Segment StartTime="2017.987" EndTime="2021.763" Participant="mn007"> |
| <Pause/> placing the decision boundaries at <Pause/> optimal places for |
| </Segment> |
| <Segment StartTime="2021.763" EndTime="2023.639" Participant="mn007"> |
| <Pause/> one kind of <Pause/> data <Pause/> but |
| </Segment> |
| <Segment StartTime="2023.639" EndTime="2025.902" Participant="mn007"> |
| <Pause/> this is not the case for another kind of data. |
| </Segment> |
| <Segment StartTime="2025.902" EndTime="2028.214" Participant="me013"> |
| <Emphasis> It's </Emphasis> a <Emphasis> trade-off, </Emphasis> right? <Emphasis> Any- </Emphasis> anyway <Emphasis> go </Emphasis> ahead. |
| </Segment> |
| <Segment StartTime="2026.129" EndTime="2026.562" Participant="mn007"> |
| So - |
| </Segment> |
| <Segment StartTime="2027.217" EndTime="2030.769" Participant="mn007"> |
| <Pause/> Yeah. So uh what we were thinking about is perhaps |
| </Segment> |
| <Segment StartTime="2030.769" EndTime="2033.137" Participant="mn007"> |
| <Pause/> um one way |
| </Segment> |
| <Segment StartTime="2033.137" EndTime="2038.449" Participant="mn007"> |
| <Pause/> to solve this problem is increase the number of <Pause/> outputs of the neural networks. |
| </Segment> |
| <Segment StartTime="2040.009" EndTime="2041.903" Participant="mn007"> |
| Doing something like, um |
| </Segment> |
| <Segment StartTime="2041.903" EndTime="2046.218" Participant="mn007"> |
| <Pause/> um phonemes within context and, |
| </Segment> |
| <Segment StartTime="2046.218" EndTime="2049.047" Participant="mn007"> |
| well, basically context dependent phonemes. |
| </Segment> |
| <Segment StartTime="2052.976" EndTime="2054.826" Participant="me013"> |
| Maybe. I mean, I - I think |
| </Segment> |
| <Segment StartTime="2054.826" EndTime="2057.130" Participant="me013"> |
| <Pause/> you could make <Pause/> the same argument, |
| </Segment> |
| <Segment StartTime="2057.130" EndTime="2058.549" Participant="me013"> |
| <Pause/> it'd be just as legitimate, |
| </Segment> |
| <Segment StartTime="2058.549" EndTime="2059.579" Participant="me013"> |
| <Pause/> for hybrid systems |
| </Segment> |
| <Segment StartTime="2059.579" EndTime="2060.157" Participant="me013"> |
| <Pause/> as well. |
| </Segment> |
| <Segment StartTime="2060.988" EndTime="2062.897" Participant="mn007"> |
| Yeah but, we know that - |
| </Segment> |
| <Segment StartTime="2061.259" EndTime="2062.073" Participant="me013"> |
| Right. |
| </Segment> |
| <Segment StartTime="2062.073" EndTime="2063.008" Participant="me013"> |
| <Pause/> And in fact, |
| </Segment> |
| <Segment StartTime="2063.008" EndTime="2066.074" Participant="me013"> |
| <Pause/> th- things get <Emphasis> better </Emphasis> with context <Emphasis> dependent </Emphasis> <Pause/> <Emphasis> versions. </Emphasis> |
| </Segment> |
| <Segment StartTime="2067.465" EndTime="2068.040" Participant="me013"> |
| Right? |
| </Segment> |
| <Segment StartTime="2068.040" EndTime="2072.283" Participant="mn007"> |
| Ye- yeah but here it's something different. We want to have features uh well, |
| </Segment> |
| <Segment StartTime="2071.560" EndTime="2072.347" Participant="me013"> |
| Yeah. |
| </Segment> |
| <Segment StartTime="2072.283" EndTime="2074.484" Participant="mn007"> |
| <Pause/> um. |
| </Segment> |
| <Segment StartTime="2074.556" EndTime="2075.796" Participant="me013"> |
| <Emphasis> Yeah, </Emphasis> but it's <Emphasis> still </Emphasis> true |
| </Segment> |
| <Segment StartTime="2075.796" EndTime="2076.971" Participant="me013"> |
| <Pause/> that what you're doing |
| </Segment> |
| <Segment StartTime="2076.971" EndTime="2079.428" Participant="me013"> |
| <Pause/> is <Pause/> you're ignoring - |
| </Segment> |
| <Segment StartTime="2079.428" EndTime="2081.654" Participant="me013"> |
| <Pause/> you're - you're coming up with something to represent, |
| </Segment> |
| <Segment StartTime="2081.654" EndTime="2083.009" Participant="me013"> |
| <Pause/> whether it's a distribution, |
| </Segment> |
| <Segment StartTime="2083.009" EndTime="2084.863" Participant="me013"> |
| <Pause/> probability distribution or <Emphasis> features, </Emphasis> |
| </Segment> |
| <Segment StartTime="2084.530" EndTime="2087.147" Participant="me034"> |
| <VocalSound Description="sips coffee"/> |
| </Segment> |
| <Segment StartTime="2084.863" EndTime="2087.032" Participant="me013"> |
| <Pause/> you're coming up with a set of variables |
| </Segment> |
| <Segment StartTime="2087.032" EndTime="2088.152" Participant="me013"> |
| <Pause/> that are representing |
| </Segment> |
| <Segment StartTime="2088.152" EndTime="2089.543" Participant="me013"> |
| <Pause/> uh, |
| </Segment> |
| <Segment StartTime="2089.543" EndTime="2091.862" Participant="me013"> |
| <Pause/> things that vary w- over context. |
| </Segment> |
| <Segment StartTime="2091.558" EndTime="2092.284" Participant="mn007"> |
| Mm-hmm. |
| </Segment> |
| <Segment StartTime="2092.690" EndTime="2094.776" Participant="me013"> |
| Uh, and you're <Pause/> putting it all <Emphasis> together, </Emphasis> |
| </Segment> |
| <Segment StartTime="2094.776" EndTime="2096.709" Participant="me013"> |
| <Pause/> <Emphasis> ignoring </Emphasis> the differences in context. |
| </Segment> |
| <Segment StartTime="2097.089" EndTime="2098.278" Participant="me013"> |
| That - that's true |
| </Segment> |
| <Segment StartTime="2098.278" EndTime="2099.435" Participant="me013"> |
| <Pause/> for the hybrid system, |
| </Segment> |
| <Segment StartTime="2099.435" EndTime="2100.801" Participant="me013"> |
| <Pause/> it's true for a tandem system. |
| </Segment> |
| <Segment StartTime="2101.434" EndTime="2103.186" Participant="me013"> |
| So, for that reason, when you - in - |
| </Segment> |
| <Segment StartTime="2103.186" EndTime="2104.325" Participant="me013"> |
| in - in a <Emphasis> hybrid </Emphasis> system, |
| </Segment> |
| <Segment StartTime="2104.325" EndTime="2106.656" Participant="me013"> |
| <Pause/> when you incorporate <Emphasis> context </Emphasis> one way or another, |
| </Segment> |
| <Segment StartTime="2106.656" EndTime="2108.170" Participant="me013"> |
| <Pause/> you <Emphasis> do </Emphasis> get better <Emphasis> scores. </Emphasis> |
| </Segment> |
| <Segment StartTime="2108.903" EndTime="2109.416" Participant="mn007"> |
| Yeah. |
| </Segment> |
| <Segment StartTime="2109.579" EndTime="2113.224" Participant="me013"> |
| O_K? <Pause/> But I - it's - it's a big <Emphasis> deal </Emphasis> |
| </Segment> |
| <Segment StartTime="2113.224" EndTime="2114.597" Participant="me013"> |
| <Pause/> to <Emphasis> get </Emphasis> that. |
| </Segment> |
| <Segment StartTime="2115.537" EndTime="2117.380" Participant="me013"> |
| I - I'm - I'm sort of - |
| </Segment> |
| <Segment StartTime="2118.826" EndTime="2123.452" Participant="me013"> |
| And <Emphasis> once </Emphasis> you - the other thing is that <Emphasis> once </Emphasis> you represent - start representing more and more <Emphasis> context </Emphasis> |
| </Segment> |
| <Segment StartTime="2123.452" EndTime="2125.797" Participant="me013"> |
| <Pause/> it <Emphasis> is </Emphasis> <Pause/> uh |
| </Segment> |
| <Segment StartTime="2125.797" EndTime="2127.116" Participant="me013"> |
| <Pause/> much more |
| </Segment> |
| <Segment StartTime="2127.116" EndTime="2129.488" Participant="me013"> |
| <Pause/> um <Emphasis> specific </Emphasis> |
| </Segment> |
| <Segment StartTime="2129.488" EndTime="2131.977" Participant="me013"> |
| <Pause/> to a particular task in <Emphasis> language. </Emphasis> |
| </Segment> |
| <Segment StartTime="2132.573" EndTime="2134.123" Participant="me013"> |
| So um |
| </Segment> |
| <Segment StartTime="2135.680" EndTime="2137.304" Participant="me013"> |
| Uh, the - |
| </Segment> |
| <Segment StartTime="2137.304" EndTime="2140.228" Participant="me013"> |
| <Pause/> the <Emphasis> acoustics </Emphasis> <Emphasis> associated </Emphasis> with <Pause/> uh |
| </Segment> |
| <Segment StartTime="2138.487" EndTime="2140.106" Participant="me006"> |
| <NonVocalSound Description="mike noise"/> |
| </Segment> |
| <Segment StartTime="2140.228" EndTime="2144.442" Participant="me013"> |
| a particular context, for instance you may have some kinds of contexts that will <Emphasis> never </Emphasis> occur |
| </Segment> |
| <Segment StartTime="2144.442" EndTime="2149.583" Participant="me013"> |
| <Pause/> in one language and will occur <Emphasis> frequently </Emphasis> in the <Emphasis> other, </Emphasis> so the qu- the issue of getting enough <Emphasis> training </Emphasis> |
| </Segment> |
| <Segment StartTime="2149.583" EndTime="2152.637" Participant="me013"> |
| <Pause/> for a particular kind of <Emphasis> context </Emphasis> becomes harder. |
| </Segment> |
| <Segment StartTime="2153.430" EndTime="2156.195" Participant="me013"> |
| We already actually don't have a huge amount of training data |
| </Segment> |
| <Segment StartTime="2156.195" EndTime="2158.815" Participant="me013"> |
| <Pause/> um |
| </Segment> |
| <Segment StartTime="2158.815" EndTime="2159.480" Participant="me013"> |
| <VocalSound Description="inbreath"/> |
| </Segment> |
| <Segment StartTime="2159.282" EndTime="2160.298" Participant="mn007"> |
| Yeah, but - |
| </Segment> |
| <Segment StartTime="2162.817" EndTime="2163.840" Participant="mn007"> |
| mmm, |
| </Segment> |
| <Segment StartTime="2165.040" EndTime="2170.303" Participant="mn007"> |
| I mean, <Pause/> the - the way we - we do it now is that we have a neural network and |
| </Segment> |
| <Segment StartTime="2166.017" EndTime="2167.191" Participant="me006"> |
| <NonVocalSound Description="mike noise"/> |
| </Segment> |
| <Segment StartTime="2170.303" EndTime="2171.423" Participant="mn007"> |
| <Pause/> basically |
| </Segment> |
| <Segment StartTime="2171.423" EndTime="2175.893" Participant="mn007"> |
| <Pause/> the net- network is trained almost to give binary decisions. |
| </Segment> |
| <Segment StartTime="2176.563" EndTime="2177.132" Participant="me013"> |
| Right. |
| </Segment> |
| <Segment StartTime="2177.700" EndTime="2180.139" Participant="mn007"> |
| And <Pause/> uh - |
| </Segment> |
| <Segment StartTime="2180.139" EndTime="2182.484" Participant="mn007"> |
| binary decisions about <Emphasis> phonemes. </Emphasis> Nnn - |
| </Segment> |
| <Segment StartTime="2183.870" EndTime="2188.280" Participant="me018"> |
| <VocalSound Description="yawn"/> |
| </Segment> |
| <Segment StartTime="2184.706" EndTime="2185.577" Participant="mn007"> |
| Uh |
| </Segment> |
| <Segment StartTime="2184.979" EndTime="2185.785" Participant="me013"> |
| <Emphasis> Almost. </Emphasis> |
| </Segment> |
| <Segment StartTime="2185.785" EndTime="2188.542" Participant="me013"> |
| But I mean it - it - it <Emphasis> does </Emphasis> give a <Emphasis> distribution. </Emphasis> |
| </Segment> |
| <Segment StartTime="2186.082" EndTime="2186.697" Participant="mn007"> |
| It's - |
| </Segment> |
| <Segment StartTime="2188.970" EndTime="2189.366" Participant="mn007"> |
| Yeah. |
| </Segment> |
| <Segment StartTime="2189.544" EndTime="2190.560" Participant="me013"> |
| It's - and - |
| </Segment> |
| <Segment StartTime="2190.560" EndTime="2194.798" Participant="me013"> |
| and <Pause/> it <Emphasis> is </Emphasis> true that if there's two phones that are very <Emphasis> similar, </Emphasis> |
| </Segment> |
| <Segment StartTime="2194.798" EndTime="2197.198" Participant="me013"> |
| <Pause/> that <Pause/> uh <Pause/> the - |
| </Segment> |
| <Segment StartTime="2197.198" EndTime="2199.185" Participant="me013"> |
| <Pause/> i- it may prefer one but it will |
| </Segment> |
| <Segment StartTime="2199.185" EndTime="2201.787" Participant="me013"> |
| <Pause/> give a reasonably high value to the other, too. |
| </Segment> |
| <Segment StartTime="2200.991" EndTime="2201.552" Participant="mn007"> |
| Yeah. |
| </Segment> |
| <Segment StartTime="2202.560" EndTime="2207.330" Participant="mn007"> |
| Yeah, sure but <VocalSound Description="inbreath"/> uh |
| </Segment> |
| <Segment StartTime="2203.062" EndTime="2204.519" Participant="me006"> |
| <NonVocalSound Description="mike noise"/> |
| </Segment> |
| <Segment StartTime="2207.330" EndTime="2210.836" Participant="mn007"> |
| <Pause/> So basically it's almost binary decisions and |
| </Segment> |
| <Segment StartTime="2210.836" EndTime="2215.109" Participant="mn007"> |
| <Pause/> um the idea of using more <Pause/> classes is |
| </Segment> |
| <Segment StartTime="2215.109" EndTime="2220.508" Participant="mn007"> |
| <Pause/> to <Pause/> get something that's <Pause/> less binary decisions. |
| </Segment> |
| <Segment StartTime="2219.920" EndTime="2222.380" Participant="me013"> |
| Oh no, but it would <Emphasis> still </Emphasis> be even <Emphasis> more </Emphasis> of a binary decision. |
| </Segment> |
| <Segment StartTime="2222.380" EndTime="2223.531" Participant="me013"> |
| <Pause/> It - it'd be even <Emphasis> more </Emphasis> of one. |
| </Segment> |
| <Segment StartTime="2223.531" EndTime="2224.869" Participant="me013"> |
| <Pause/> Because then you would say |
| </Segment> |
| <Segment StartTime="2224.441" EndTime="2225.814" Participant="mn007"> |
| But - yeah, but - |
| </Segment> |
| <Segment StartTime="2224.869" EndTime="2228.356" Participant="me013"> |
| <Pause/> that in - that this phone in <Emphasis> this </Emphasis> context is a <Emphasis> one, </Emphasis> |
| </Segment> |
| <Segment StartTime="2228.356" EndTime="2231.518" Participant="me013"> |
| <Pause/> but the <Emphasis> same </Emphasis> phone in a slightly <Emphasis> different </Emphasis> context is a <Emphasis> zero. </Emphasis> |
| </Segment> |
| <Segment StartTime="2232.210" EndTime="2235.108" Participant="me013"> |
| That would be even - even <Emphasis> more </Emphasis> distinct of a binary decision. |
| </Segment> |
| <Segment StartTime="2235.108" EndTime="2238.216" Participant="me013"> |
| <Emphasis> I </Emphasis> actually would have thought you'd wanna go the <Emphasis> other </Emphasis> way and have <Emphasis> fewer </Emphasis> classes. |
| </Segment> |
| <Segment StartTime="2235.363" EndTime="2236.182" Participant="mn007"> |
| Yeah, but if - |
| </Segment> |
| <Segment StartTime="2238.980" EndTime="2240.558" Participant="me013"> |
| Uh, I mean for instance, |
| </Segment> |
| <Segment StartTime="2240.558" EndTime="2242.329" Participant="me013"> |
| the - the thing I was arguing for <Emphasis> before, </Emphasis> but |
| </Segment> |
| <Segment StartTime="2241.071" EndTime="2241.804" Participant="mn007"> |
| Mmm. |
| </Segment> |
| <Segment StartTime="2242.329" EndTime="2244.010" Participant="me013"> |
| again which I don't think we have time to <Emphasis> try, </Emphasis> |
| </Segment> |
| <Segment StartTime="2244.010" EndTime="2249.792" Participant="me013"> |
| <Pause/> is something in which you would modify the code so you could train to have several outputs on and use articulatory features |
| </Segment> |
| <Segment StartTime="2249.643" EndTime="2250.569" Participant="mn007"> |
| Mm-hmm. |
| </Segment> |
| <Segment StartTime="2249.792" EndTime="2251.690" Participant="me013"> |
| <Pause/> cuz then that would - that would go - |
| </Segment> |
| <Segment StartTime="2251.690" EndTime="2254.654" Participant="me013"> |
| <Pause/> that would be much <Emphasis> broader </Emphasis> and cover many different <Emphasis> situations. </Emphasis> |
| </Segment> |
| <Segment StartTime="2254.654" EndTime="2258.413" Participant="me013"> |
| <Pause/> But if you go to <Emphasis> very </Emphasis> very <Emphasis> fine </Emphasis> categories, it's <Emphasis> very </Emphasis> <Pause/> binary. |
| </Segment> |
| <Segment StartTime="2254.785" EndTime="2257.739" Participant="mn007"> |
| Mmm. <Pause/> Yeah, but I think - |
| </Segment> |
| <Segment StartTime="2258.680" EndTime="2264.332" Participant="mn007"> |
| Yeah, perhaps you're right, but you have more classes so <Pause/> you - you have more information in your features. So, |
| </Segment> |
| <Segment StartTime="2264.332" EndTime="2264.750" Participant="mn007"> |
| <VocalSound Description="inbreath"/> |
| </Segment> |
| <Segment StartTime="2265.166" EndTime="2265.771" Participant="me013"> |
| Mm-hmm. |
| </Segment> |
| <Segment StartTime="2266.740" EndTime="2269.980" Participant="mn007"> |
| Um <Pause/> You have more information in the |
| </Segment> |
| <Segment StartTime="2267.168" EndTime="2267.866" Participant="me013"> |
| True. |
| </Segment> |
| <Segment StartTime="2269.980" EndTime="2272.256" Participant="mn007"> |
| <Pause/> uh posteriors vector |
| </Segment> |
| <Segment StartTime="2272.256" EndTime="2273.991" Participant="mn007"> |
| <Pause/> um |
| </Segment> |
| <Segment StartTime="2273.991" EndTime="2275.382" Participant="mn007"> |
| <Pause/> which means that - |
| </Segment> |
| <Segment StartTime="2276.590" EndTime="2281.804" Participant="mn007"> |
| But still the information is relevant because it's - it's information that helps to discriminate, |
| </Segment> |
| <Segment StartTime="2277.396" EndTime="2278.183" Participant="me013"> |
| Mm-hmm. |
| </Segment> |
| <Segment StartTime="2281.804" EndTime="2283.846" Participant="mn007"> |
| <Pause/> if it's possible to be able to discriminate |
| </Segment> |
| <Segment StartTime="2282.078" EndTime="2282.671" Participant="me013"> |
| Mm-hmm. |
| </Segment> |
| <Segment StartTime="2283.846" EndTime="2286.828" Participant="mn007"> |
| <Pause/> among the phonemes in context. |
| </Segment> |
| <Segment StartTime="2287.071" EndTime="2288.209" Participant="me013"> |
| Well it's - it's - |
| </Segment> |
| <Segment StartTime="2287.476" EndTime="2288.330" Participant="mn007"> |
| But the - |
| </Segment> |
| <Segment StartTime="2288.209" EndTime="2291.639" Participant="me013"> |
| <Pause/> it's an interesting thought. I mean we - we could disagree about it at length |
| </Segment> |
| <Segment StartTime="2289.670" EndTime="2290.534" Participant="mn007"> |
| Mmm. |
| </Segment> |
| <Segment StartTime="2291.314" EndTime="2292.013" Participant="mn007"> |
| Mmm. |
| </Segment> |
| <Segment StartTime="2291.639" EndTime="2294.724" Participant="me013"> |
| but the - the <Emphasis> real </Emphasis> thing is if you're interested in it you'll probably try it and - |
| </Segment> |
| <Segment StartTime="2294.724" EndTime="2295.903" Participant="me013"> |
| <Pause/> and <Pause/> we'll see. |
| </Segment> |
| <Segment StartTime="2294.742" EndTime="2295.448" Participant="mn007"> |
| <VocalSound Description="sniff"/> |
| </Segment> |
| <Segment StartTime="2295.448" EndTime="2296.075" Participant="mn007"> |
| <VocalSound Description="clear throat"/> |
| </Segment> |
| <Segment StartTime="2295.903" EndTime="2300.886" Participant="me013"> |
| <Pause/> But - but what I'm <Emphasis> more </Emphasis> concerned with now, as an operational level, is |
| </Segment> |
| <Segment StartTime="2298.749" EndTime="2299.437" Participant="mn007"> |
| Mmm. |
| </Segment> |
| <Segment StartTime="2300.886" EndTime="2303.522" Participant="me013"> |
| <Pause/> uh, you know, what do we do in four or five days? |
| </Segment> |
| <Segment StartTime="2303.503" EndTime="2304.243" Participant="mn007"> |
| <VocalSound Description="inbreath"/> |
| </Segment> |
| <Segment StartTime="2303.522" EndTime="2305.383" Participant="me013"> |
| <Pause/> Uh, and - |
| </Segment> |
| <Segment StartTime="2305.383" EndTime="2308.760" Participant="me013"> |
| <Pause/> so we have <Pause/> to be concerned <Pause/> with |
| </Segment> |
| <Segment StartTime="2308.760" EndTime="2314.217" Participant="me013"> |
| <Pause/> Are we gonna look at any combinations of things, you know once the nets get retrained so you have this problem out of it. |
| </Segment> |
| <Segment StartTime="2312.826" EndTime="2313.578" Participant="mn007"> |
| Mmm. |
| </Segment> |
| <Segment StartTime="2314.217" EndTime="2319.567" Participant="me013"> |
| <Pause/> Um, are we going to look at <Pause/> multi-band? Are we gonna look at combinations of things? |
| </Segment> |
| <Segment StartTime="2319.567" EndTime="2322.114" Participant="me013"> |
| <Pause/> Uh, what <Emphasis> questions </Emphasis> are we gonna ask, |
| </Segment> |
| <Segment StartTime="2322.114" EndTime="2323.451" Participant="me013"> |
| <Pause/> uh now that, I mean, |
| </Segment> |
| <Segment StartTime="2323.451" EndTime="2325.760" Participant="me013"> |
| <Pause/> we should probably turn shortly to this O_G_<Emphasis> I </Emphasis> note. |
| </Segment> |
| <Segment StartTime="2325.760" EndTime="2328.521" Participant="me013"> |
| <Pause/> Um, how are we going to <Pause/> combine |
| </Segment> |
| <Segment StartTime="2328.521" EndTime="2330.325" Participant="me013"> |
| <Pause/> with what <Emphasis> they've </Emphasis> been focusing on? |
| </Segment> |
| <Segment StartTime="2330.325" EndTime="2331.755" Participant="me013"> |
| <Pause/> Uh, <Pause/> |
| </Segment> |
| <Segment StartTime="2331.755" EndTime="2334.809" Participant="me013"> |
| Uh we haven't been doing any of the L_D_<Emphasis> A </Emphasis> RASTA sort of thing. |
| </Segment> |
| <Segment StartTime="2333.189" EndTime="2333.812" Participant="mn007"> |
| Mm-hmm. |
| </Segment> |
| <Segment StartTime="2334.809" EndTime="2338.208" Participant="me013"> |
| <Pause/> And they, although they don't <Emphasis> talk </Emphasis> about it in this <Emphasis> note, </Emphasis> um, |
| </Segment> |
| <Segment StartTime="2338.208" EndTime="2341.554" Participant="me013"> |
| <Pause/> there's um, <Pause/> the issue of the |
| </Segment> |
| <Segment StartTime="2341.554" EndTime="2348.219" Participant="me013"> |
| <Pause/> um <Uncertain> Mu law </Uncertain> <Pause/> business <Pause/> uh <Pause/> versus the logarithm, |
| </Segment> |
| <Segment StartTime="2348.219" EndTime="2350.530" Participant="me013"> |
| <Pause/> um, <Pause/> so. |
| </Segment> |
| <Segment StartTime="2348.476" EndTime="2349.180" Participant="mn007"> |
| Mm-hmm. |
| </Segment> |
| <Segment StartTime="2351.001" EndTime="2353.916" Participant="me013"> |
| So what i- what is going on right <Emphasis> now? </Emphasis> What's right - you've got |
| </Segment> |
| <Segment StartTime="2352.596" EndTime="2354.088" Participant="mn007"> |
| <VocalSound Description="inbreath"/> |
| </Segment> |
| <Segment StartTime="2353.916" EndTime="2355.154" Participant="me013"> |
| <Pause/> nets retraining, |
| </Segment> |
| <Segment StartTime="2355.154" EndTime="2359.220" Participant="me013"> |
| <Pause/> Are there - is there - are there any H_T_<Emphasis> K </Emphasis> <Pause/> trainings - testings going on? |
| </Segment> |
| <Segment StartTime="2355.826" EndTime="2356.431" Participant="mn007"> |
| N- |
| </Segment> |
| <Segment StartTime="2358.127" EndTime="2362.582" Participant="fn002"> |
| I - I - I'm trying the H_T_K with eh, |
| </Segment> |
| <Segment StartTime="2362.582" EndTime="2368.852" Participant="fn002"> |
| <Pause/> P_L_P twelve on-line delta-delta and <Pronounce Pronunciation=""M_S_..G""> M_S_G </Pronounce> filter <Pause/> together. |
| </Segment> |
| <Segment StartTime="2368.852" EndTime="2369.972" Participant="me013"> |
| The combination, I see. |
| </Segment> |
| <Segment StartTime="2369.648" EndTime="2373.251" Participant="fn002"> |
| The combination, yeah. But I haven't result <VocalSound Description="laugh"/> at this moment. |
| </Segment> |
| <Segment StartTime="2373.413" EndTime="2375.200" Participant="me013"> |
| M_S_G and - and P_L_P. |
| </Segment> |
| <Segment StartTime="2375.076" EndTime="2375.347" Participant="fn002"> |
| Yeah. |
| </Segment> |
| <Segment StartTime="2375.200" EndTime="2379.320" Participant="me013"> |
| <Pause/> And is this with the revised <Pause/> on-line normalization? |
| </Segment> |
| <Segment StartTime="2379.160" EndTime="2381.673" Participant="fn002"> |
| Ye- Uh, with the old <Pause/> <Uncertain> older </Uncertain>, yeah. |
| </Segment> |
| <Segment StartTime="2379.872" EndTime="2380.517" Participant="mn007"> |
| Yeah. |
| </Segment> |
| <Segment StartTime="2380.748" EndTime="2381.420" Participant="me013"> |
| <Emphasis> Old </Emphasis> one. |
| </Segment> |
| <Segment StartTime="2381.420" EndTime="2384.937" Participant="me013"> |
| <Pause/> So it's using all the nets for that but again we have the hope that it - |
| </Segment> |
| <Segment StartTime="2383.119" EndTime="2385.070" Participant="fn002"> |
| Yeah. <Pause/> But <Pause/> We can |
| </Segment> |
| <Segment StartTime="2384.252" EndTime="2384.877" Participant="mn007"> |
| <VocalSound Description="sniff"/> |
| </Segment> |
| <Segment StartTime="2384.937" EndTime="2386.570" Participant="me013"> |
| <Pause/> We have the hope that it - |
| </Segment> |
| <Segment StartTime="2385.757" EndTime="2386.606" Participant="fn002"> |
| know soon. |
| </Segment> |
| <Segment StartTime="2386.570" EndTime="2388.125" Participant="me013"> |
| <Pause/> maybe it's not making too much difference, |
| </Segment> |
| <Segment StartTime="2387.636" EndTime="2388.503" Participant="fn002"> |
| Maybe. <VocalSound Description="laugh"/> |
| </Segment> |
| <Segment StartTime="2387.974" EndTime="2388.887" Participant="mn007"> |
| <VocalSound Description="inbreath"/> |
| </Segment> |
| <Segment StartTime="2388.125" EndTime="2388.902" Participant="me013"> |
| <Pause/> but - but yeah. |
| </Segment> |
| <Segment StartTime="2388.706" EndTime="2389.273" Participant="fn002"> |
| I don't know. |
| </Segment> |
| <Segment StartTime="2389.540" EndTime="2389.799" Participant="mn007"> |
| Yeah. |
| </Segment> |
| <Segment StartTime="2390.510" EndTime="2391.271" Participant="me013"> |
| Uh, O_K. |
| </Segment> |
| <Segment StartTime="2391.280" EndTime="2395.415" Participant="mn007"> |
| Uh so there is this combination, yeah. Working on combination obviously. |
| </Segment> |
| <Segment StartTime="2395.415" EndTime="2399.225" Participant="mn007"> |
| <Pause/> Um, I will start work on multi-band. |
| </Segment> |
| <Segment StartTime="2395.605" EndTime="2396.115" Participant="fn002"> |
| Mm-hmm. |
| </Segment> |
| <Segment StartTime="2399.225" EndTime="2403.128" Participant="mn007"> |
| <Pause/> And <Pause/> we |
| </Segment> |
| <Segment StartTime="2403.128" EndTime="2407.154" Participant="mn007"> |
| <Pause/> plan to work also on the idea of using both |
| </Segment> |
| <Segment StartTime="2407.154" EndTime="2410.009" Participant="mn007"> |
| <Pause/> features <Pause/> and net outputs. |
| </Segment> |
| <Segment StartTime="2410.490" EndTime="2410.841" Participant="fn002"> |
| Yep. |
| </Segment> |
| <Segment StartTime="2411.180" EndTime="2412.672" Participant="mn007"> |
| Um. |
| </Segment> |
| <Segment StartTime="2413.550" EndTime="2415.868" Participant="mn007"> |
| And <Pause/> we think that |
| </Segment> |
| <Segment StartTime="2415.868" EndTime="2417.874" Participant="mn007"> |
| <Pause/> with this approach perhaps |
| </Segment> |
| <Segment StartTime="2417.874" EndTime="2421.334" Participant="mn007"> |
| <Pause/> we could reduce the number of outputs of the neural network. |
| </Segment> |
| <Segment StartTime="2421.334" EndTime="2423.270" Participant="mn007"> |
| <Pause/> Um, |
| </Segment> |
| <Segment StartTime="2423.270" EndTime="2425.102" Participant="mn007"> |
| So, get simpler networks, |
| </Segment> |
| <Segment StartTime="2425.102" EndTime="2426.692" Participant="mn007"> |
| <Pause/> because we still have the features. |
| </Segment> |
| <Segment StartTime="2426.692" EndTime="2429.276" Participant="mn007"> |
| <Pause/> So we have um |
| </Segment> |
| <Segment StartTime="2429.276" EndTime="2431.805" Participant="mn007"> |
| <Pause/> come up with um |
| </Segment> |
| <Segment StartTime="2431.805" EndTime="2437.051" Participant="mn007"> |
| <Pause/> different kind of <Pause/> broad phonetic categories. |
| </Segment> |
| <Segment StartTime="2437.925" EndTime="2438.827" Participant="mn007"> |
| And we have - |
| </Segment> |
| <Segment StartTime="2438.827" EndTime="2442.278" Participant="mn007"> |
| Basically we have three <Pause/> types of broad phonetic classes. |
| </Segment> |
| <Segment StartTime="2442.278" EndTime="2450.334" Participant="mn007"> |
| <Pause/> Well, something using place of articulation which - which leads to <Pause/> nine, I think, <Pause/> broad classes. |
| </Segment> |
| <Segment StartTime="2450.334" EndTime="2455.917" Participant="mn007"> |
| <Pause/> Uh, another which is based on manner, which is - is also something like nine classes. |
| </Segment> |
| <Segment StartTime="2455.917" EndTime="2458.754" Participant="mn007"> |
| <Pause/> And then, <Pause/> something that combine both, |
| </Segment> |
| <Segment StartTime="2458.754" EndTime="2461.526" Participant="mn007"> |
| <Pause/> and we have <Pause/> twenty f- <Pause/> twenty-five? |
| </Segment> |
| <Segment StartTime="2461.823" EndTime="2462.710" Participant="me006"> |
| Twenty-seven. |
| </Segment> |
| <Segment StartTime="2462.710" EndTime="2464.559" Participant="mn007"> |
| Twenty-seven broad classes. |
| </Segment> |
| <Segment StartTime="2465.832" EndTime="2467.946" Participant="mn007"> |
| So like, uh, <Pause/> oh, I don't know, |
| </Segment> |
| <Segment StartTime="2467.946" EndTime="2470.403" Participant="mn007"> |
| <Pause/> like back vowels, front vowels. |
| </Segment> |
| <Segment StartTime="2470.176" EndTime="2470.468" Participant="me013"> |
| <VocalSound Description="inbreath"/> |
| </Segment> |
| <Segment StartTime="2470.468" EndTime="2471.454" Participant="me013"> |
| So what you do - |
| </Segment> |
| <Segment StartTime="2471.057" EndTime="2471.965" Participant="mn007"> |
| Um |
| </Segment> |
| <Segment StartTime="2471.454" EndTime="2472.652" Participant="me013"> |
| <Pause/> um I just wanna understand so |
| </Segment> |
| <Segment StartTime="2472.652" EndTime="2475.379" Participant="me013"> |
| <Pause/> You have <Emphasis> two </Emphasis> net or <Emphasis> three </Emphasis> nets? <Uncertain> Was this? </Uncertain> |
| </Segment> |
| <Segment StartTime="2475.959" EndTime="2477.487" Participant="me013"> |
| How many - how many nets do you have? |
| </Segment> |
| <Segment StartTime="2477.890" EndTime="2479.804" Participant="mn007"> |
| For the moments we do not - don't have nets, I mean, |
| </Segment> |
| <Segment StartTime="2479.448" EndTime="2480.580" Participant="me013"> |
| No nets. |
| </Segment> |
| <Segment StartTime="2479.804" EndTime="2481.460" Participant="mn007"> |
| <Pause/> It's just - |
| </Segment> |
| <Segment StartTime="2481.460" EndTime="2483.054" Participant="fn002"> |
| Begin to work in this. |
| </Segment> |
| <Segment StartTime="2481.460" EndTime="2485.390" Participant="mn007"> |
| Were we just changing <Pause/> the labels to retrain nets <Pause/> with fewer out- outputs. |
| </Segment> |
| <Segment StartTime="2483.896" EndTime="2484.949" Participant="fn002"> |
| <Uncertain> We are </Uncertain> <Uncertain NumSyllables="2"> @@ </Uncertain>. <Comment Description="whispered"/> |
| </Segment> |
| <Segment StartTime="2485.920" EndTime="2486.200" Participant="me013"> |
| Right. |
| </Segment> |
| <Segment StartTime="2486.750" EndTime="2488.680" Participant="me013"> |
| But - but I didn't understand - |
| </Segment> |
| <Segment StartTime="2486.750" EndTime="2487.686" Participant="mn007"> |
| And then - |
| </Segment> |
| <Segment StartTime="2488.499" EndTime="2489.134" Participant="mn007"> |
| Mm-hmm. |
| </Segment> |
| <Segment StartTime="2490.050" EndTime="2490.820" Participant="me013"> |
| Uh. |
| </Segment> |
| <Segment StartTime="2490.820" EndTime="2495.904" Participant="me013"> |
| <Pause/> the software currently just has - uh a - allows for I think, the one - one hot |
| </Segment> |
| <Segment StartTime="2494.932" EndTime="2495.185" Participant="mn007"> |
| <VocalSound Description="sniff"/> |
| </Segment> |
| <Segment StartTime="2496.485" EndTime="2499.680" Participant="me013"> |
| output. So you're having multiple nets and combining them, <Pause/> or - ? <Comment Description="leading question"/> |
| </Segment> |
| <Segment StartTime="2501.060" EndTime="2502.855" Participant="me013"> |
| Uh, how are you - how are you coming up with - |
| </Segment> |
| <Segment StartTime="2503.282" EndTime="2510.442" Participant="me013"> |
| If you say <Pause/> uh <Pause/> If you have a <Emphasis> place </Emphasis> <Pause/> characteristic and a <Emphasis> manner </Emphasis> characteristic, how do you - |
| </Segment> |
| <Segment StartTime="2509.163" EndTime="2509.808" Participant="mn007"> |
| <VocalSound Description="sniff"/> |
| </Segment> |
| <Segment StartTime="2510.272" EndTime="2510.715" Participant="mn007"> |
| It- |
| </Segment> |
| <Segment StartTime="2511.650" EndTime="2513.779" Participant="me018"> |
| I think they have one output. |
| </Segment> |
| <Segment StartTime="2511.650" EndTime="2513.247" Participant="mn007"> |
| It's the single net, yeah. |
| </Segment> |
| <Segment StartTime="2513.737" EndTime="2514.820" Participant="me013"> |
| Oh, it's just one net. |
| </Segment> |
| <Segment StartTime="2514.341" EndTime="2515.437" Participant="me006"> |
| mm-hmm |
| </Segment> |
| <Segment StartTime="2514.433" EndTime="2515.025" Participant="fn002"> |
| Yeah. |
| </Segment> |
| <Segment StartTime="2514.658" EndTime="2516.194" Participant="mn007"> |
| It's one net with |
| </Segment> |
| <Segment StartTime="2516.194" EndTime="2520.784" Participant="mn007"> |
| <Pause/> um <Pause/> twenty-seven outputs if we have twenty-seven classes, yeah. |
| </Segment> |
| <Segment StartTime="2519.830" EndTime="2520.690" Participant="me013"> |
| I see. |
| </Segment> |
| <Segment StartTime="2521.050" EndTime="2522.025" Participant="me013"> |
| I see, O_K. |
| </Segment> |
| <Segment StartTime="2522.045" EndTime="2526.529" Participant="mn007"> |
| So it's - Well, it's basically a standard net with fewer <Pause/> classes. |
| </Segment> |
| <Segment StartTime="2527.200" EndTime="2530.725" Participant="me013"> |
| So you're sort of going the <Emphasis> other </Emphasis> way of what you were saying a bit ago instead of - yeah. |
| </Segment> |
| <Segment StartTime="2530.051" EndTime="2531.425" Participant="mn007"> |
| Yeah, but I think - Yeah. |
| </Segment> |
| <Segment StartTime="2531.303" EndTime="2532.904" Participant="me006"> |
| But <Emphasis> including </Emphasis> the <Emphasis> features. </Emphasis> |
| </Segment> |
| <Segment StartTime="2531.733" EndTime="2532.261" Participant="mn007"> |
| B- b- |
| </Segment> |
| <Segment StartTime="2531.944" EndTime="2533.103" Participant="fn002"> |
| Yeah. <VocalSound Description="laugh"/> |
| </Segment> |
| <Segment StartTime="2532.261" EndTime="2533.611" Participant="mn007"> |
| <Emphasis> including </Emphasis> the features, yeah. |
| </Segment> |
| <Segment StartTime="2533.611" EndTime="2536.058" Participant="mn007"> |
| I don't think this <Pause/> will work <Pause/> alone. |
| </Segment> |
| <Segment StartTime="2536.058" EndTime="2537.684" Participant="mn007"> |
| I think it will get worse because |
| </Segment> |
| <Segment StartTime="2538.493" EndTime="2539.239" Participant="me013"> |
| Uh-huh. |
| </Segment> |
| <Segment StartTime="2538.600" EndTime="2541.420" Participant="mn007"> |
| Well, I believe the effect that - of - of too |
| </Segment> |
| <Segment StartTime="2542.440" EndTime="2544.263" Participant="mn007"> |
| reducing too much the information is |
| </Segment> |
| <Segment StartTime="2544.263" EndTime="2547.534" Participant="mn007"> |
| <Pause/> basically - basically what happens and - |
| </Segment> |
| <Segment StartTime="2547.534" EndTime="2548.596" Participant="mn007"> |
| <Pause/> but - |
| </Segment> |
| <Segment StartTime="2547.644" EndTime="2548.019" Participant="me013"> |
| <VocalSound Description="inbreath"/> |
| </Segment> |
| <Segment StartTime="2548.019" EndTime="2549.203" Participant="me013"> |
| But you think if you include that |
| </Segment> |
| <Segment StartTime="2549.203" EndTime="2550.672" Participant="me013"> |
| <Pause/> <Emphasis> plus </Emphasis> the other features, <Comment Description="providing lead-in"/> |
| </Segment> |
| <Segment StartTime="2550.042" EndTime="2553.486" Participant="mn007"> |
| Yeah, because <Pause/> there is perhaps one important thing that the net |
| </Segment> |
| <Segment StartTime="2553.486" EndTime="2556.034" Participant="mn007"> |
| <Pause/> brings, and O_G_I <Emphasis> show- </Emphasis> <Emphasis> showed </Emphasis> that, is |
| </Segment> |
| <Segment StartTime="2556.034" EndTime="2559.500" Participant="mn007"> |
| <Pause/> the distinction between <Pause/> sp- speech and <Emphasis> silence </Emphasis> |
| </Segment> |
| <Segment StartTime="2560.070" EndTime="2565.918" Participant="mn007"> |
| Because these nets are trained on well-controlled condition. I mean the labels are obtained on clean speech, and we add noise after. |
| </Segment> |
| <Segment StartTime="2565.918" EndTime="2567.417" Participant="mn007"> |
| <Pause/> So this is one thing |
| </Segment> |
| <Segment StartTime="2567.417" EndTime="2569.320" Participant="mn007"> |
| <Pause/> And |
| </Segment> |
| <Segment StartTime="2570.140" EndTime="2573.339" Participant="mn007"> |
| But perhaps, something intermediary using also |
| </Segment> |
| <Segment StartTime="2573.339" EndTime="2576.628" Participant="mn007"> |
| <Pause/> some broad classes could - could bring so much more information. |
| </Segment> |
| <Segment StartTime="2576.628" EndTime="2578.489" Participant="mn007"> |
| <Pause/> Uh. |
| </Segment> |
| <Segment StartTime="2578.830" EndTime="2581.898" Participant="me013"> |
| <VocalSound Description="inbreath"/> So - so again then we have these broad classes |
| </Segment> |
| <Segment StartTime="2581.898" EndTime="2582.444" Participant="me013"> |
| <Pause/> and - |
| </Segment> |
| <Segment StartTime="2582.444" EndTime="2585.130" Participant="me013"> |
| well, <Emphasis> somewhat </Emphasis> broad. I mean, it's twenty-seven instead of sixty-four, |
| </Segment> |
| <Segment StartTime="2585.130" EndTime="2585.752" Participant="me013"> |
| <Pause/> basically. |
| </Segment> |
| <Segment StartTime="2585.984" EndTime="2586.462" Participant="mn007"> |
| Yeah. |
| </Segment> |
| <Segment StartTime="2586.469" EndTime="2588.737" Participant="me013"> |
| And you have the <Emphasis> original </Emphasis> features. |
| </Segment> |
| <Segment StartTime="2589.060" EndTime="2589.536" Participant="mn007"> |
| Yeah. |
| </Segment> |
| <Segment StartTime="2589.090" EndTime="2590.189" Participant="me013"> |
| Which are P_L_P, or something. |
| </Segment> |
| <Segment StartTime="2590.175" EndTime="2590.753" Participant="mn007"> |
| Mm-hmm. |
| </Segment> |
| <Segment StartTime="2590.590" EndTime="2594.063" Participant="me013"> |
| And then uh, just to remind me, all of that goes <Pause/> into - |
| </Segment> |
| <Segment StartTime="2594.583" EndTime="2599.760" Participant="me013"> |
| uh, that all of that is transformed by uh, uh, K_- K_L or something, or - ? <Comment Description="leading question"/> |
| </Segment> |
| <Segment StartTime="2599.499" EndTime="2604.672" Participant="mn007"> |
| There will probably be, yeah, one single K_L to transform everything or |
| </Segment> |
| <Segment StartTime="2600.151" EndTime="2600.743" Participant="fn002"> |
| Mu. |
| </Segment> |
| <Segment StartTime="2600.426" EndTime="2600.912" Participant="me006"> |
| <NonVocalSound Description="mike noise"/> |
| </Segment> |
| <Segment StartTime="2603.746" EndTime="2606.985" Participant="me006"> |
| <NonVocalSound Description="mike noise"/> |
| </Segment> |
| <Segment StartTime="2604.534" EndTime="2605.047" Participant="me013"> |
| Right. |
| </Segment> |
| <Segment StartTime="2604.672" EndTime="2605.332" Participant="mn007"> |
| <VocalSound Description="inbreath"/> |
| </Segment> |
| <Segment StartTime="2605.332" EndTime="2607.111" Participant="mn007"> |
| <Pause/> uh, per- |
| </Segment> |
| <Segment StartTime="2605.940" EndTime="2608.714" Participant="fn002"> |
| No transform the P_L_P and only |
| </Segment> |
| <Segment StartTime="2609.099" EndTime="2610.276" Participant="fn002"> |
| transform the other |
| </Segment> |
| <Segment StartTime="2609.698" EndTime="2612.168" Participant="me006"> |
| <NonVocalSound Description="mike noise"/> |
| </Segment> |
| <Segment StartTime="2611.414" EndTime="2612.261" Participant="fn002"> |
| I'm not sure. |
| </Segment> |
| <Segment StartTime="2612.478" EndTime="2612.822" Participant="me013"> |
| <VocalSound Description="inbreath"/> |
| </Segment> |
| <Segment StartTime="2612.660" EndTime="2615.782" Participant="mn007"> |
| This is <Pause/> still something <Pause/> that yeah, we <Pause/> don't know - |
| </Segment> |
| <Segment StartTime="2612.822" EndTime="2613.513" Participant="me013"> |
| Well no, I think - |
| </Segment> |
| <Segment StartTime="2614.576" EndTime="2615.127" Participant="me013"> |
| <Emphasis> I </Emphasis> see. |
| </Segment> |
| <Segment StartTime="2614.638" EndTime="2616.379" Participant="me006"> |
| <NonVocalSound Description="mike noise"/> |
| </Segment> |
| <Segment StartTime="2615.082" EndTime="2615.949" Participant="fn002"> |
| Two e- |
| </Segment> |
| <Segment StartTime="2615.127" EndTime="2616.830" Participant="me013"> |
| So there's a question of whether you would - |
| </Segment> |
| <Segment StartTime="2615.782" EndTime="2616.640" Participant="mn007"> |
| Yeah. |
| </Segment> |
| <Segment StartTime="2616.830" EndTime="2617.451" Participant="me013"> |
| <VocalSound Description="inbreath"/> |
| </Segment> |
| <Segment StartTime="2618.247" EndTime="2618.896" Participant="me013"> |
| Right. |
| </Segment> |
| <Segment StartTime="2619.390" EndTime="2621.430" Participant="me013"> |
| Whether you would transform <Emphasis> together </Emphasis> or just <Emphasis> one. </Emphasis> |
| </Segment> |
| <Segment StartTime="2621.241" EndTime="2622.306" Participant="fn002"> |
| <Uncertain NumSyllables="1"> @@ </Uncertain> it's one. |
| </Segment> |
| <Segment StartTime="2621.430" EndTime="2621.817" Participant="me013"> |
| Yeah. |
| </Segment> |
| <Segment StartTime="2621.817" EndTime="2623.537" Participant="me013"> |
| <Pause/> Might wanna try it both ways. |
| </Segment> |
| <Segment StartTime="2623.537" EndTime="2624.347" Participant="me013"> |
| But that's <Emphasis> interesting. </Emphasis> |
| </Segment> |
| <Segment StartTime="2624.638" EndTime="2626.703" Participant="me006"> |
| <NonVocalSound Description="mike noise"/> |
| </Segment> |
| <Segment StartTime="2626.480" EndTime="2629.606" Participant="me013"> |
| So that's something that you're - you <Emphasis> haven't </Emphasis> trained yet but are <Emphasis> preparing </Emphasis> to train, and - |
| </Segment> |
| <Segment StartTime="2629.749" EndTime="2630.052" Participant="mn007"> |
| Yeah. |
| </Segment> |
| <Segment StartTime="2630.471" EndTime="2630.989" Participant="me013"> |
| Yeah. |
| </Segment> |
| <Segment StartTime="2631.964" EndTime="2632.663" Participant="mn007"> |
| <VocalSound Description="sniff"/> |
| </Segment> |
| <Segment StartTime="2632.780" EndTime="2633.153" Participant="me013"> |
| Um <Pause/> |
| </Segment> |
| <Segment StartTime="2633.129" EndTime="2633.879" Participant="mn007"> |
| Mmm. |
| </Segment> |
| <Segment StartTime="2633.153" EndTime="2634.856" Participant="me013"> |
| <Pause/> Yeah, so I think Hynek will be here Monday. |
| </Segment> |
| <Segment StartTime="2634.856" EndTime="2638.374" Participant="me013"> |
| <Pause/> Monday or Tuesday. So <VocalSound Description="laugh"/> |
| </Segment> |
| <Segment StartTime="2635.592" EndTime="2635.957" Participant="mn007"> |
| <VocalSound Description="laugh"/> |
| </Segment> |
| <Segment StartTime="2636.795" EndTime="2637.679" Participant="fn002"> |
| <VocalSound Description="laugh"/> |
| </Segment> |
| <Segment StartTime="2638.277" EndTime="2639.577" Participant="mn007"> |
| Uh, yeah. |
| </Segment> |
| <Segment StartTime="2638.374" EndTime="2639.999" Participant="me013"> |
| <Pause/> <VocalSound Description="laugh"/> |
| </Segment> |
| <Segment StartTime="2641.096" EndTime="2642.194" Participant="me013"> |
| So I think, you know, we need to |
| </Segment> |
| <Segment StartTime="2642.194" EndTime="2644.217" Participant="me013"> |
| <Pause/> choose the - choose the experiments carefully, |
| </Segment> |
| <Segment StartTime="2644.217" EndTime="2646.568" Participant="me013"> |
| <Pause/> so we can get uh key - |
| </Segment> |
| <Segment StartTime="2644.517" EndTime="2647.796" Participant="me006"> |
| <NonVocalSound Description="mike noise"/> |
| </Segment> |
| <Segment StartTime="2646.568" EndTime="2647.985" Participant="me013"> |
| <Pause/> key questions answered |
| </Segment> |
| <Segment StartTime="2647.637" EndTime="2647.863" Participant="mn007"> |
| <VocalSound Description="sniff"/> |
| </Segment> |
| <Segment StartTime="2647.863" EndTime="2648.455" Participant="mn007"> |
| Mm-hmm. |
| </Segment> |
| <Segment StartTime="2647.985" EndTime="2649.429" Participant="me013"> |
| <Pause/> uh <Emphasis> before </Emphasis> then and |
| </Segment> |
| <Segment StartTime="2649.429" EndTime="2652.108" Participant="me013"> |
| <Pause/> leave other ones <Emphasis> aside </Emphasis> even if it |
| </Segment> |
| <Segment StartTime="2651.925" EndTime="2652.436" Participant="mn007"> |
| <VocalSound Description="sniff"/> |
| </Segment> |
| <Segment StartTime="2652.108" EndTime="2656.024" Participant="me013"> |
| <Pause/> leaves incomplete <Pause/> tables <VocalSound Description="laugh"/> <Pause/> someplace, uh |
| </Segment> |
| <Segment StartTime="2654.153" EndTime="2656.623" Participant="me006"> |
| <NonVocalSound Description="mike noise"/> |
| </Segment> |
| <Segment StartTime="2656.024" EndTime="2658.280" Participant="me013"> |
| <Pause/> uh, it's - it's really time to - |
| </Segment> |
| <Segment StartTime="2658.280" EndTime="2659.688" Participant="me013"> |
| <Pause/> time to choose. |
| </Segment> |
| <Segment StartTime="2658.327" EndTime="2658.954" Participant="mn007"> |
| Mm-hmm. |
| </Segment> |
| <Segment StartTime="2660.470" EndTime="2661.853" Participant="me013"> |
| Um, let me pass this out, |
| </Segment> |
| <Segment StartTime="2661.853" EndTime="2662.562" Participant="me013"> |
| <Pause/> by the way. |
| </Segment> |
| <Segment StartTime="2662.562" EndTime="2663.348" Participant="me013"> |
| <Pause/> Um |
| </Segment> |
| <Segment StartTime="2665.120" EndTime="2665.691" Participant="me013"> |
| These are - |
| </Segment> |
| <Segment StartTime="2665.368" EndTime="2665.696" Participant="mn007"> |
| <VocalSound Description="sniff"/> |
| </Segment> |
| <Segment StartTime="2665.696" EndTime="2666.144" Participant="mn007"> |
| <VocalSound Description="sniff"/> |
| </Segment> |
| <Segment StartTime="2666.659" EndTime="2667.169" Participant="me013"> |
| Did - did - |
| </Segment> |
| <Segment StartTime="2667.169" EndTime="2669.701" Participant="me013"> |
| <Pause/> did I interrupt you? Were there other things that you wanted to - |
| </Segment> |
| <Segment StartTime="2667.241" EndTime="2669.207" Participant="fn002"> |
| Yeah, I have one. <Comment Description="whispered"/> |
| </Segment> |
| <Segment StartTime="2669.133" EndTime="2670.064" Participant="me006"> |
| <NonVocalSound Description="mike noise"/> |
| </Segment> |
| <Segment StartTime="2670.106" EndTime="2671.726" Participant="mn007"> |
| Uh, no. I don't think so. |
| </Segment> |
| <Segment StartTime="2673.761" EndTime="2676.745" Participant="fn002"> |
| <Uncertain NumSyllables="4"> @@ </Uncertain> <Comment Description="whispered"/> |
| </Segment> |
| <Segment StartTime="2679.333" EndTime="2680.249" Participant="mn007"> |
| Yeah, I have one. <Comment Description="whispered"/> |
| </Segment> |
| <Segment StartTime="2680.121" EndTime="2682.362" Participant="me026"> |
| Oh, thanks. |
| </Segment> |
| <Segment StartTime="2681.531" EndTime="2682.583" Participant="me013"> |
| Ah! <Pause/> O_K. |
| </Segment> |
| <Segment StartTime="2681.562" EndTime="2690.833" Participant="me006"> |
| <NonVocalSound Description="mike noise"/> |
| </Segment> |
| <Segment StartTime="2681.907" EndTime="2684.166" Participant="fn002"> |
| We have one. <Uncertain NumSyllables="3"> @@ </Uncertain> <VocalSound Description="laugh"/> |
| </Segment> |
| <Segment StartTime="2682.583" EndTime="2684.078" Participant="me013"> |
| <Pause/> O_K, we have <Pause/> lots of them. |
| </Segment> |
| <Segment StartTime="2684.078" EndTime="2688.045" Participant="me013"> |
| <Pause/> O_K, so <VocalSound Description="swallow"/> um, |
| </Segment> |
| <Segment StartTime="2688.045" EndTime="2688.678" Participant="me013"> |
| Something I asked - |
| </Segment> |
| <Segment StartTime="2688.678" EndTime="2691.651" Participant="me013"> |
| So they're - they're doing <Pause/> the - the V_A_D <Pause/> I guess they mean |
| </Segment> |
| <Segment StartTime="2691.651" EndTime="2693.084" Participant="me013"> |
| voice <Emphasis> activity </Emphasis> detection |
| </Segment> |
| <Segment StartTime="2693.350" EndTime="2694.730" Participant="me013"> |
| So again, it's the silence - |
| </Segment> |
| <Segment StartTime="2694.730" EndTime="2696.310" Participant="me013"> |
| So they've just trained up a net |
| </Segment> |
| <Segment StartTime="2696.310" EndTime="2699.031" Participant="me013"> |
| <Pause/> which has two outputs, I believe. |
| </Segment> |
| <Segment StartTime="2699.031" EndTime="2699.692" Participant="me013"> |
| <Pause/> Um |
| </Segment> |
| <Segment StartTime="2699.692" EndTime="2700.630" Participant="me013"> |
| <VocalSound Description="inbreath"/> |
| </Segment> |
| <Segment StartTime="2700.502" EndTime="2701.242" Participant="mn007"> |
| <VocalSound Description="sniff"/> |
| </Segment> |
| <Segment StartTime="2700.630" EndTime="2702.896" Participant="me013"> |
| I asked uh <Pause/> Hynek whether - |
| </Segment> |
| <Segment StartTime="2702.896" EndTime="2705.083" Participant="me013"> |
| <Pause/> I haven't talked to Sunil - I asked Hynek whether |
| </Segment> |
| <Segment StartTime="2705.083" EndTime="2707.365" Participant="me013"> |
| <Pause/> they compared that to |
| </Segment> |
| <Segment StartTime="2707.365" EndTime="2711.615" Participant="me013"> |
| <Pause/> just taking the nets we already <Emphasis> had </Emphasis> <Pause/> and summing up the probabilities. |
| </Segment> |
| <Segment StartTime="2710.887" EndTime="2711.411" Participant="mn007"> |
| Mm-hmm. |
| </Segment> |
| <Segment StartTime="2712.230" EndTime="2713.360" Participant="me013"> |
| Uh. <Pause/> |
| </Segment> |
| <Segment StartTime="2713.360" EndTime="2714.395" Participant="me013"> |
| To get the speech - |
| </Segment> |
| <Segment StartTime="2714.395" EndTime="2716.729" Participant="me013"> |
| voice activity detection, or else just using the silence, |
| </Segment> |
| <Segment StartTime="2716.729" EndTime="2718.313" Participant="me013"> |
| <Pause/> if there's only <Emphasis> one </Emphasis> <Pause/> silence output. |
| </Segment> |
| <Segment StartTime="2718.313" EndTime="2722.445" Participant="me013"> |
| <Pause/> Um <Pause/> And, he didn't think they <Emphasis> had, </Emphasis> um. |
| </Segment> |
| <Segment StartTime="2723.420" EndTime="2726.027" Participant="me013"> |
| But on the other hand, maybe they can get by with a smaller net |
| </Segment> |
| <Segment StartTime="2726.027" EndTime="2731.131" Participant="me013"> |
| <Pause/> and <Pause/> maybe <Pause/> sometimes you don't <Emphasis> run </Emphasis> the other, maybe there's a <Emphasis> computational </Emphasis> advantage to having a separate net, anyway. |
| </Segment> |
| <Segment StartTime="2729.881" EndTime="2730.465" Participant="mn007"> |
| Mm-hmm. |
| </Segment> |
| <Segment StartTime="2732.220" EndTime="2733.774" Participant="me013"> |
| <VocalSound Description="inbreath"/> So um |
| </Segment> |
| <Segment StartTime="2734.621" EndTime="2735.407" Participant="me013"> |
| Their uh - |
| </Segment> |
| <Segment StartTime="2735.407" EndTime="2736.720" Participant="me013"> |
| <Pause/> the <Emphasis> results </Emphasis> look pretty <Emphasis> good. </Emphasis> |
| </Segment> |
| <Segment StartTime="2738.505" EndTime="2739.674" Participant="mn007"> |
| Yeah. <VocalSound Description="outbreath"/> |
| </Segment> |
| <Segment StartTime="2738.620" EndTime="2740.634" Participant="me013"> |
| Um, <Pause/> I mean, not <Emphasis> uniformly. </Emphasis> |
| </Segment> |
| <Segment StartTime="2740.634" EndTime="2745.526" Participant="me013"> |
| <Pause/> I mean, there's a - an example or two <Pause/> that you can find, where it made it slightly worse, but |
| </Segment> |
| <Segment StartTime="2745.526" EndTime="2747.635" Participant="me013"> |
| <Pause/> uh in - in <Emphasis> all </Emphasis> but a <Emphasis> couple </Emphasis> |
| </Segment> |
| <Segment StartTime="2746.566" EndTime="2747.219" Participant="mn007"> |
| Mmm. |
| </Segment> |
| <Segment StartTime="2747.635" EndTime="2749.416" Participant="me013"> |
| <Pause/> examples. |
| </Segment> |
| <Segment StartTime="2749.234" EndTime="2750.250" Participant="fn002"> |
| <VocalSound Description="inbreath"/> |
| </Segment> |
| <Segment StartTime="2749.416" EndTime="2750.738" Participant="me013"> |
| <Pause/> Uh. |
| </Segment> |
| <Segment StartTime="2750.250" EndTime="2757.890" Participant="fn002"> |
| But they have a question of the result. Um how are <Uncertain> trained </Uncertain> the - the L_D_A filter? |
| </Segment> |
| <Segment StartTime="2750.577" EndTime="2751.027" Participant="mn007"> |
| <VocalSound Description="sniff"/> |
| </Segment> |
| <Segment StartTime="2760.257" EndTime="2763.000" Participant="fn002"> |
| How <Uncertain> obtained </Uncertain> the L_D_A filter? |
| </Segment> |
| <Segment StartTime="2760.855" EndTime="2761.472" Participant="mn007"> |
| Mmm. |
| </Segment> |
| <Segment StartTime="2761.924" EndTime="2762.374" Participant="mn007"> |
| <VocalSound Description="sniff"/> |
| </Segment> |
| <Segment StartTime="2763.000" EndTime="2764.665" Participant="me013"> |
| I- I'm sorry. I don't understand your question. |
| </Segment> |
| <Segment StartTime="2763.687" EndTime="2766.423" Participant="fn002"> |
| Yes, um the L_D_A filter |
| </Segment> |
| <Segment StartTime="2766.423" EndTime="2769.386" Participant="fn002"> |
| <Pause/> needs some <Pause/> training set |
| </Segment> |
| <Segment StartTime="2769.386" EndTime="2771.247" Participant="fn002"> |
| <Pause/> to obtain the filter. |
| </Segment> |
| <Segment StartTime="2771.247" EndTime="2771.820" Participant="fn002"> |
| <Pause/> Maybe |
| </Segment> |
| <Segment StartTime="2772.580" EndTime="2775.720" Participant="fn002"> |
| I don't know exactly <Emphasis> how </Emphasis> <Pause/> they are obtained. |
| </Segment> |
| <Segment StartTime="2774.517" EndTime="2776.091" Participant="me013"> |
| It's on <Pause/> <Emphasis> training. </Emphasis> |
| </Segment> |
| <Segment StartTime="2776.000" EndTime="2778.240" Participant="fn002"> |
| Training, with the training test of each - |
| </Segment> |
| <Segment StartTime="2779.778" EndTime="2780.263" Participant="me013"> |
| <VocalSound Description="inbreath"/> |
| </Segment> |
| <Segment StartTime="2781.024" EndTime="2782.067" Participant="fn002"> |
| You understand me? |
| </Segment> |
| <Segment StartTime="2782.691" EndTime="2783.507" Participant="me013"> |
| No. |
| </Segment> |
| <Segment StartTime="2783.240" EndTime="2785.316" Participant="fn002"> |
| Yeah, uh for example, |
| </Segment> |
| <Segment StartTime="2785.316" EndTime="2786.780" Participant="fn002"> |
| <Pause/> L_D_A filter |
| </Segment> |
| <Segment StartTime="2786.780" EndTime="2790.158" Participant="fn002"> |
| <Pause/> need a set of - |
| </Segment> |
| <Segment StartTime="2790.158" EndTime="2792.452" Participant="fn002"> |
| <Pause/> a set of training <Pause/> to obtain the filter. |
| </Segment> |
| <Segment StartTime="2792.452" EndTime="2805.183" Participant="fn002"> |
| <Pause/> And maybe <Pause/> for the Italian, for the <Uncertain> T_D </Uncertain> <Pause/> <Uncertain> T_E </Uncertain> on for Finnish, these filter are - are obtained with their own training set. |
| </Segment> |
| <Segment StartTime="2792.737" EndTime="2793.456" Participant="me013"> |
| Yes. |
| </Segment> |
| <Segment StartTime="2806.410" EndTime="2807.371" Participant="me013"> |
| Yes, I don't know. |
| </Segment> |
| <Segment StartTime="2807.800" EndTime="2808.773" Participant="me013"> |
| That's - that's - so that's a - |
| </Segment> |
| <Segment StartTime="2808.773" EndTime="2810.078" Participant="me013"> |
| that's a very good question, then - |
| </Segment> |
| <Segment StartTime="2810.078" EndTime="2810.432" Participant="me013"> |
| now that it - |
| </Segment> |
| <Segment StartTime="2810.432" EndTime="2811.426" Participant="me013"> |
| <Pause/> I understand it. It's |
| </Segment> |
| <Segment StartTime="2811.426" EndTime="2813.690" Participant="me013"> |
| <Pause/> "yeah, where does the L_D_A come from?" In the - |
| </Segment> |
| <Segment StartTime="2813.690" EndTime="2816.601" Participant="me013"> |
| In <Pause/> <Emphasis> earlier </Emphasis> experiments, they had taken L_D_A |
| </Segment> |
| <Segment StartTime="2816.601" EndTime="2819.323" Participant="me013"> |
| <Pause/> from a <Emphasis> completely </Emphasis> different <Emphasis> database, </Emphasis> right? |
| </Segment> |
| <Segment StartTime="2818.260" EndTime="2819.047" Participant="fn002"> |
| Yeah. |
| </Segment> |
| <Segment StartTime="2819.336" EndTime="2824.410" Participant="fn002"> |
| Yeah, because maybe it the same situation that the neural network training with their own |
| </Segment> |
| <Segment StartTime="2824.090" EndTime="2824.811" Participant="mn007"> |
| Mmm. |
| </Segment> |
| <Segment StartTime="2825.640" EndTime="2826.117" Participant="fn002"> |
| set. |
| </Segment> |
| <Segment StartTime="2826.371" EndTime="2828.790" Participant="me013"> |
| <Pause/> So that's a good question. <Pause/> <Emphasis> Where </Emphasis> does it come from? <Pause/> |
| </Segment> |
| <Segment StartTime="2828.790" EndTime="2830.269" Participant="me013"> |
| Yeah, I don't know. |
| </Segment> |
| <Segment StartTime="2830.269" EndTime="2830.865" Participant="me013"> |
| Um, <Pause/> |
| </Segment> |
| <Segment StartTime="2830.865" EndTime="2838.544" Participant="me013"> |
| but uh to tell you the <Pause/> truth, I wasn't actually looking at the L_D_A so much when I - I was looking at it I was <Pause/> mostly thinking about the - <Pause/> the V_A_D. |
| </Segment> |
| <Segment StartTime="2839.273" EndTime="2840.628" Participant="me013"> |
| And um, it ap- |
| </Segment> |
| <Segment StartTime="2840.628" EndTime="2841.328" Participant="me013"> |
| <Pause/> it ap- |
| </Segment> |
| <Segment StartTime="2841.328" EndTime="2843.841" Participant="me013"> |
| <Pause/> Oh what does - what does A_S_P? |
| </Segment> |
| <Segment StartTime="2845.400" EndTime="2845.900" Participant="me013"> |
| Oh that's - |
| </Segment> |
| <Segment StartTime="2845.900" EndTime="2847.855" Participant="mn007"> |
| The features, yeah. |
| </Segment> |
| <Segment StartTime="2848.635" EndTime="2849.055" Participant="mn007"> |
| Yeah. |
| </Segment> |
| <Segment StartTime="2848.810" EndTime="2851.134" Participant="fn002"> |
| I don't understand also what is - |
| </Segment> |
| <Segment StartTime="2849.525" EndTime="2851.306" Participant="me013"> |
| It says "baseline A_S_P". |
| </Segment> |
| <Segment StartTime="2851.134" EndTime="2855.323" Participant="fn002"> |
| <Pause/> what is the <Emphasis> difference </Emphasis> between A_S_P and uh baseline <Uncertain> over </Uncertain>? |
| </Segment> |
| <Segment StartTime="2855.253" EndTime="2856.314" Participant="mn007"> |
| Yeah, I don't know. |
| </Segment> |
| <Segment StartTime="2855.720" EndTime="2856.566" Participant="me034"> |
| A_S_P. |
| </Segment> |
| <Segment StartTime="2857.890" EndTime="2858.483" Participant="fn002"> |
| This is - |
| </Segment> |
| <Segment StartTime="2858.741" EndTime="2859.330" Participant="me034"> |
| Oh. |
| </Segment> |
| <Segment StartTime="2859.110" EndTime="2860.442" Participant="me013"> |
| Anybody know <Pause/> <Uncertain> any </Uncertain> - |
| </Segment> |
| <Segment StartTime="2859.510" EndTime="2860.081" Participant="me034"> |
| <Emphasis> There </Emphasis> it is. |
| </Segment> |
| <Segment StartTime="2865.990" EndTime="2866.490" Participant="me013"> |
| Um |
| </Segment> |
| <Segment StartTime="2867.221" EndTime="2869.710" Participant="me013"> |
| Cuz there's "baseline <Emphasis> Aurora </Emphasis>" <Pause/> <Emphasis> above </Emphasis> it. |
| </Segment> |
| <Segment StartTime="2868.864" EndTime="2869.957" Participant="me034"> |
| <VocalSound Description="laugh"/> Mm-hmm. |
| </Segment> |
| <Segment StartTime="2870.550" EndTime="2871.740" Participant="me013"> |
| And it's - |
| </Segment> |
| <Segment StartTime="2872.080" EndTime="2876.616" Participant="me013"> |
| This is <Emphasis> mostly </Emphasis> better than baseline, although in some cases it's a little worse, in a couple cases. |
| </Segment> |
| <Segment StartTime="2875.858" EndTime="2878.141" Participant="me034"> |
| Well, it says baseline A_S_P is twenty-three mill |
| </Segment> |
| <Segment StartTime="2876.402" EndTime="2877.112" Participant="fn002"> |
| Yeah. |
| </Segment> |
| <Segment StartTime="2878.141" EndTime="2879.100" Participant="me034"> |
| <Pause/> minus thirteen. |
| </Segment> |
| <Segment StartTime="2879.135" EndTime="2882.052" Participant="me013"> |
| Yeah, it <Emphasis> says </Emphasis> what it <Emphasis> is. </Emphasis> But I don't how that's different <Pause/> from - |
| </Segment> |
| <Segment StartTime="2882.220" EndTime="2883.440" Participant="me034"> |
| From the baseline. <Comment Description="completing [me013]'s question"/> O_K. |
| </Segment> |
| <Segment StartTime="2883.770" EndTime="2884.607" Participant="me013"> |
| I think this was - |
| </Segment> |
| <Segment StartTime="2884.273" EndTime="2884.823" Participant="fn002"> |
| Yeah. <Comment Description="whispered"/> |
| </Segment> |
| <Segment StartTime="2884.607" EndTime="2886.413" Participant="me013"> |
| <Pause/> I think this is the same point we were at |
| </Segment> |
| <Segment StartTime="2885.124" EndTime="2886.745" Participant="mn007"> |
| I think - <Pause/> |
| </Segment> |
| <Segment StartTime="2886.413" EndTime="2887.692" Participant="me013"> |
| when - when we were up in Oregon. |
| </Segment> |
| <Segment StartTime="2886.745" EndTime="2891.867" Participant="mn007"> |
| I think it's the C_zero - using C_zero instead of log energy. Yeah, it's <Emphasis> this. </Emphasis> |
| </Segment> |
| <Segment StartTime="2889.965" EndTime="2891.800" Participant="fn002"> |
| Ah, O_K, mm-hmm. |
| </Segment> |
| <Segment StartTime="2892.370" EndTime="2893.135" Participant="fn002"> |
| <VocalSound Description="inbreath"/> yeah. |
| </Segment> |
| <Segment StartTime="2893.270" EndTime="2893.660" Participant="me013"> |
| Oh. |
| </Segment> |
| <Segment StartTime="2894.050" EndTime="2894.779" Participant="me013"> |
| O_K. |
| </Segment> |
| <Segment StartTime="2894.050" EndTime="2895.238" Participant="mn007"> |
| It should be that, yeah. |
| </Segment> |
| <Segment StartTime="2895.979" EndTime="2896.503" Participant="mn007"> |
| Because - |
| </Segment> |
| <Segment StartTime="2896.000" EndTime="2896.867" Participant="me013"> |
| Shouldn't it be - |
| </Segment> |
| <Segment StartTime="2896.108" EndTime="2896.426" Participant="me018"> |
| They s- |
| </Segment> |
| <Segment StartTime="2896.589" EndTime="2901.522" Participant="me018"> |
| they say in here that the V_A_D is <Emphasis> not </Emphasis> used as an additional feature. Does - does anybody know <Emphasis> how </Emphasis> they're using it? |
| </Segment> |
| <Segment StartTime="2901.800" EndTime="2904.054" Participant="me013"> |
| <Emphasis> Yeah. </Emphasis> So - so what they're doing here is, |
| </Segment> |
| <Segment StartTime="2902.177" EndTime="2902.759" Participant="mn007"> |
| Yeah. |
| </Segment> |
| <Segment StartTime="2904.054" EndTime="2905.698" Participant="me013"> |
| <Pause/> i- if you look down at the block diagram, |
| </Segment> |
| <Segment StartTime="2905.698" EndTime="2907.599" Participant="me013"> |
| <Pause/> um, <Pause/> |
| </Segment> |
| <Segment StartTime="2907.599" EndTime="2909.238" Participant="me013"> |
| they estimate - they get a - |
| </Segment> |
| <Segment StartTime="2907.645" EndTime="2908.171" Participant="me018"> |
| But that - |
| </Segment> |
| <Segment StartTime="2909.238" EndTime="2911.960" Participant="me013"> |
| <Pause/> they get an estimate <Pause/> of whether it's speech or silence, |
| </Segment> |
| <Segment StartTime="2911.960" EndTime="2913.680" Participant="me013"> |
| <Pause/> and then they have a median <Emphasis> filter </Emphasis> of it. |
| </Segment> |
| <Segment StartTime="2913.718" EndTime="2914.240" Participant="me018"> |
| Mm-hmm. |
| </Segment> |
| <Segment StartTime="2914.560" EndTime="2915.908" Participant="me013"> |
| And so um, |
| </Segment> |
| <Segment StartTime="2915.908" EndTime="2918.129" Participant="me013"> |
| <Pause/> basically they're trying to find <Emphasis> stretches. </Emphasis> |
| </Segment> |
| <Segment StartTime="2918.129" EndTime="2921.818" Participant="me013"> |
| <Pause/> The median filter is enforcing a - i- it having some <Emphasis> continuity. </Emphasis> |
| </Segment> |
| <Segment StartTime="2921.775" EndTime="2922.549" Participant="me018"> |
| Mm-hmm. |
| </Segment> |
| <Segment StartTime="2921.818" EndTime="2927.051" Participant="me013"> |
| <Pause/> You find <Emphasis> stretches </Emphasis> where the <Pause/> combination of the <Pause/> frame wise V_A_D and the - |
| </Segment> |
| <Segment StartTime="2927.051" EndTime="2929.851" Participant="me013"> |
| <Pause/> the median <Emphasis> filter </Emphasis> say that there's a stretch of <Emphasis> silence. </Emphasis> |
| </Segment> |
| <Segment StartTime="2929.851" EndTime="2932.608" Participant="me013"> |
| <Pause/> And then it's going through and just throwing the <Emphasis> data </Emphasis> away. |
| </Segment> |
| <Segment StartTime="2933.337" EndTime="2933.938" Participant="me034"> |
| Hmm. |
| </Segment> |
| <Segment StartTime="2934.801" EndTime="2935.989" Participant="me013"> |
| Right? |
| </Segment> |
| <Segment StartTime="2935.989" EndTime="2937.068" Participant="me013"> |
| So um - |
| </Segment> |
| <Segment StartTime="2936.735" EndTime="2938.310" Participant="me018"> |
| So it's - it's - |
| </Segment> |
| <Segment StartTime="2939.350" EndTime="2942.090" Participant="me018"> |
| I don't understand. You mean it's throwing out frames? |
| </Segment> |
| <Segment StartTime="2942.360" EndTime="2943.950" Participant="me013"> |
| It's throwing out chunks of frames, yeah. |
| </Segment> |
| <Segment StartTime="2942.360" EndTime="2943.017" Participant="me018"> |
| Before - |
| </Segment> |
| <Segment StartTime="2943.950" EndTime="2949.199" Participant="me013"> |
| There's - the - the median filter is enforcing that it's <Emphasis> not </Emphasis> gonna be single cases of frames, or isolated frames. |
| </Segment> |
| <Segment StartTime="2947.633" EndTime="2948.350" Participant="me018"> |
| Yeah. |
| </Segment> |
| <Segment StartTime="2949.199" EndTime="2950.623" Participant="me013"> |
| <Pause/> So it's throwing out frames |
| </Segment> |
| <Segment StartTime="2950.623" EndTime="2952.509" Participant="me013"> |
| <Pause/> and the thing is <Pause/> um, |
| </Segment> |
| <Segment StartTime="2952.509" EndTime="2956.242" Participant="me013"> |
| <Pause/> what I don't understand is how they're doing this with H_T_<Emphasis> K. </Emphasis> |
| </Segment> |
| <Segment StartTime="2956.242" EndTime="2956.942" Participant="me013"> |
| <Pause/> This is - |
| </Segment> |
| <Segment StartTime="2956.490" EndTime="2959.350" Participant="me018"> |
| Yeah, that's what I was just gonna ask. How can you just throw out frames? |
| </Segment> |
| <Segment StartTime="2957.892" EndTime="2958.359" Participant="me013"> |
| Yeah. |
| </Segment> |
| <Segment StartTime="2959.266" EndTime="2959.850" Participant="me013"> |
| Well, |
| </Segment> |
| <Segment StartTime="2959.707" EndTime="2960.337" Participant="mn007"> |
| i- |
| </Segment> |
| <Segment StartTime="2959.850" EndTime="2962.348" Participant="me013"> |
| you - you <Emphasis> can, </Emphasis> right? I mean y- you - you - |
| </Segment> |
| <Segment StartTime="2961.498" EndTime="2964.396" Participant="me034"> |
| <VocalSound Description="laugh"/> |
| </Segment> |
| <Segment StartTime="2962.348" EndTime="2962.910" Participant="me013"> |
| <VocalSound Description="laugh"/> |
| </Segment> |
| <Segment StartTime="2962.910" EndTime="2968.520" Participant="me013"> |
| <Pause/> it <Emphasis> stretches </Emphasis> again. For single frames I think it would be pretty <Emphasis> hard. </Emphasis> But if you say speech starts here, speech ends there. <Pause/> Right? |
| </Segment> |
| <Segment StartTime="2962.945" EndTime="2963.320" Participant="mn007"> |
| Yeah. |
| </Segment> |
| <Segment StartTime="2965.650" EndTime="2966.257" Participant="me018"> |
| Yeah. |
| </Segment> |
| <Segment StartTime="2967.957" EndTime="2968.520" Participant="me018"> |
| Mm-hmm. |
| </Segment> |
| <Segment StartTime="2969.789" EndTime="2970.236" Participant="me034"> |
| Huh. |
| </Segment> |
| <Segment StartTime="2970.244" EndTime="2970.704" Participant="mn007"> |
| Yeah. |
| </Segment> |
| <Segment StartTime="2970.507" EndTime="2971.564" Participant="me013"> |
| <VocalSound Description="laugh"/> |
| </Segment> |
| <Segment StartTime="2971.754" EndTime="2975.766" Participant="mn007"> |
| Yeah, you can basically remove the - the frames from the feature - feature files. And. |
| </Segment> |
| <Segment StartTime="2974.422" EndTime="2975.093" Participant="me013"> |
| Yeah. |
| </Segment> |
| <Segment StartTime="2976.240" EndTime="2980.827" Participant="me013"> |
| Yeah, so I mean in the - i- i- in the - in the <Emphasis> decoding, </Emphasis> you're saying that we're gonna decode from here to here. |
| </Segment> |
| <Segment StartTime="2976.687" EndTime="2977.211" Participant="mn007"> |
| I t- |
| </Segment> |
| <Segment StartTime="2981.190" EndTime="2981.580" Participant="me018"> |
| Mm-hmm. |
| </Segment> |
| <Segment StartTime="2981.734" EndTime="2983.246" Participant="me013"> |
| I think they're - they're - they're treating it, |
| </Segment> |
| <Segment StartTime="2983.246" EndTime="2984.819" Participant="me013"> |
| <Pause/> you know, like uh - |
| </Segment> |
| <Segment StartTime="2984.819" EndTime="2987.980" Participant="me013"> |
| <Pause/> well, it's not isolated word, but - but connected, you know, the - the - |
| </Segment> |
| <Segment StartTime="2988.540" EndTime="2993.540" Participant="me018"> |
| In the text they say that this - this is a <Emphasis> tentative </Emphasis> block diagram of a <Emphasis> possible </Emphasis> configuration we could <Emphasis> think </Emphasis> of. |
| </Segment> |
| <Segment StartTime="2993.991" EndTime="2995.408" Participant="me006"> |
| <NonVocalSound Description="mike noise"/> |
| </Segment> |
| <Segment StartTime="2994.925" EndTime="2996.990" Participant="me018"> |
| So that sort of sounds like they're not <Emphasis> doing </Emphasis> that yet. |
| </Segment> |
| <Segment StartTime="2998.294" EndTime="3003.929" Participant="me013"> |
| Well. <Pause/> No they - they have <Emphasis> numbers </Emphasis> though, right? <Pause/> So I think they're - they're doing something like that. I think that they're - they're - |
| </Segment> |
| <Segment StartTime="3000.773" EndTime="3001.304" Participant="mn007"> |
| <VocalSound Description="breath-laugh"/> |
| </Segment> |
| <Segment StartTime="3004.820" EndTime="3010.946" Participant="me013"> |
| I think what I mean by tha- that is they're trying to come up with a block diagram that's plausible for the standard. |
| </Segment> |
| <Segment StartTime="3011.375" EndTime="3012.597" Participant="me013"> |
| In other words, it's - uh - |
| </Segment> |
| <Segment StartTime="3012.597" EndTime="3019.837" Participant="me013"> |
| <Pause/> I mean from the point of view of - of uh reducing the number of bits you have to transmit it's not a bad idea to detect silence <Emphasis> anyway. </Emphasis> |
| </Segment> |
| <Segment StartTime="3018.706" EndTime="3019.350" Participant="me018"> |
| Yeah. |
| </Segment> |
| <Segment StartTime="3019.610" EndTime="3020.350" Participant="me018"> |
| Yeah. |
| </Segment> |
| <Segment StartTime="3020.680" EndTime="3025.250" Participant="me018"> |
| I'm just wondering what exactly <Emphasis> did </Emphasis> they do up in <Emphasis> this </Emphasis> table if it wasn't this. |
| </Segment> |
| <Segment StartTime="3020.680" EndTime="3021.284" Participant="me013"> |
| Um. |
| </Segment> |
| <Segment StartTime="3027.070" EndTime="3030.695" Participant="me013"> |
| But it's - the thing is it's that - that - that's - that's I - I - |
| </Segment> |
| <Segment StartTime="3031.304" EndTime="3034.542" Participant="me013"> |
| Certainly it would be tricky about it intrans- in transmitting <Emphasis> voice, </Emphasis> |
| </Segment> |
| <Segment StartTime="3034.542" EndTime="3038.424" Participant="me013"> |
| <Pause/> uh uh for <Emphasis> listening </Emphasis> to, is that these kinds of things |
| </Segment> |
| <Segment StartTime="3038.424" EndTime="3041.590" Participant="me013"> |
| <Pause/> uh cut <Pause/> speech <Emphasis> off </Emphasis> a lot. Right? |
| </Segment> |
| <Segment StartTime="3041.590" EndTime="3043.369" Participant="me013"> |
| <Pause/> And so <Pause/> um |
| </Segment> |
| <Segment StartTime="3041.590" EndTime="3042.212" Participant="me018"> |
| Mm-hmm. |
| </Segment> |
| <Segment StartTime="3044.360" EndTime="3045.860" Participant="me018"> |
| Plus it's gonna introduce delays. |
| </Segment> |
| <Segment StartTime="3047.030" EndTime="3050.836" Participant="me013"> |
| It <Emphasis> does </Emphasis> introduce delays but they're claiming that it's - it's within the - |
| </Segment> |
| <Segment StartTime="3050.836" EndTime="3051.873" Participant="me013"> |
| <Pause/> the boundaries of it. |
| </Segment> |
| <Segment StartTime="3051.560" EndTime="3052.030" Participant="me018"> |
| Mmm. |
| </Segment> |
| <Segment StartTime="3052.427" EndTime="3058.718" Participant="me013"> |
| <Pause/> And the L_D_A introduces delays, and b- <Pause/> what he's suggesting this here is a parallel path so that it doesn't introduce |
| </Segment> |
| <Segment StartTime="3058.718" EndTime="3060.833" Participant="me013"> |
| <Pause/> uh, any more delay. |
| </Segment> |
| <Segment StartTime="3061.220" EndTime="3066.850" Participant="me013"> |
| <Pause/> I- it introduces two hundred milliseconds of delay but at the same <Pause/> time the L_D_A <Pause/> down here - |
| </Segment> |
| <Segment StartTime="3067.432" EndTime="3070.228" Participant="me013"> |
| I <Uncertain> don't know - </Uncertain> Wh- what's the difference between <Emphasis> T_L_D_A </Emphasis> and <Emphasis> S_L_D_A </Emphasis>? |
| </Segment> |
| <Segment StartTime="3069.788" EndTime="3070.934" Participant="me034"> |
| Temporal and spectral. |
| </Segment> |
| <Segment StartTime="3071.100" EndTime="3073.106" Participant="me013"> |
| Ah, thank you. <VocalSound Description="laugh"/> |
| </Segment> |
| <Segment StartTime="3071.989" EndTime="3073.999" Participant="me034"> |
| <VocalSound Description="laugh"/> |
| </Segment> |
| <Segment StartTime="3072.358" EndTime="3073.860" Participant="fn002"> |
| <Emphasis> Temporal </Emphasis> L_D_A. |
| </Segment> |
| <Segment StartTime="3073.106" EndTime="3075.880" Participant="me013"> |
| <Pause/> Yeah, you <Emphasis> would </Emphasis> know that. <VocalSound Description="laugh"/> |
| </Segment> |
| <Segment StartTime="3073.999" EndTime="3075.278" Participant="me034"> |
| Yeah <VocalSound Description="laugh"/> |
| </Segment> |
| <Segment StartTime="3075.880" EndTime="3077.040" Participant="me013"> |
| So um. |
| </Segment> |
| <Segment StartTime="3078.294" EndTime="3081.477" Participant="me013"> |
| The <Emphasis> temporal </Emphasis> L_D_A does in fact include the same - |
| </Segment> |
| <Segment StartTime="3081.477" EndTime="3082.574" Participant="me013"> |
| <Pause/> so that - I think |
| </Segment> |
| <Segment StartTime="3082.317" EndTime="3083.410" Participant="me006"> |
| <NonVocalSound Description="mike noise"/> |
| </Segment> |
| <Segment StartTime="3082.574" EndTime="3083.594" Participant="me013"> |
| he - well, by - |
| </Segment> |
| <Segment StartTime="3083.594" EndTime="3087.371" Participant="me013"> |
| by saying this is a b- a tentative block di- diagram I think means |
| </Segment> |
| <Segment StartTime="3087.371" EndTime="3092.176" Participant="me013"> |
| <Pause/> if you construct it this way, this - this delay would work in that way and then it'd be O_K. |
| </Segment> |
| <Segment StartTime="3090.899" EndTime="3092.150" Participant="me018"> |
| <VocalSound Description="mouth"/> Ah. |
| </Segment> |
| <Segment StartTime="3092.176" EndTime="3097.072" Participant="me013"> |
| <Pause/> They - they clearly <Emphasis> did </Emphasis> actually <Emphasis> remove </Emphasis> <Pause/> <Emphasis> silent </Emphasis> sections in order - because they |
| </Segment> |
| <Segment StartTime="3097.072" EndTime="3100.340" Participant="me013"> |
| <Pause/> got these <Pause/> <Emphasis> word </Emphasis> error rate <Pause/> results. |
| </Segment> |
| <Segment StartTime="3101.346" EndTime="3102.844" Participant="me006"> |
| <NonVocalSound Description="mike noise"/> |
| </Segment> |
| <Segment StartTime="3101.685" EndTime="3103.217" Participant="me013"> |
| So um |
| </Segment> |
| <Segment StartTime="3104.180" EndTime="3108.959" Participant="me013"> |
| I think that it's - it's <Emphasis> nice </Emphasis> to do that in this because in fact, it's gonna give a better word error result |
| </Segment> |
| <Segment StartTime="3108.026" EndTime="3109.200" Participant="me006"> |
| <NonVocalSound Description="mike noise"/> |
| </Segment> |
| <Segment StartTime="3108.959" EndTime="3111.331" Participant="me013"> |
| <Pause/> and therefore will help within an evaluation. |
| </Segment> |
| <Segment StartTime="3111.331" EndTime="3114.753" Participant="me013"> |
| <Pause/> Whereas to whether this would actually be in a final standard, I don't know. |
| </Segment> |
| <Segment StartTime="3114.753" EndTime="3115.809" Participant="me013"> |
| <Pause/> Um. |
| </Segment> |
| <Segment StartTime="3116.440" EndTime="3121.314" Participant="me013"> |
| Uh, as you know, <Emphasis> part </Emphasis> of the problem with evaluation right now is that the <Pause/> word models are pretty bad and nobody wants - |
| </Segment> |
| <Segment StartTime="3121.314" EndTime="3123.207" Participant="me013"> |
| <Pause/> has - has approached improving them. |
| </Segment> |
| <Segment StartTime="3123.207" EndTime="3126.249" Participant="me013"> |
| <Pause/> So <Pause/> it's possible that a lot of the problems |
| </Segment> |
| <Segment StartTime="3126.249" EndTime="3130.227" Participant="me013"> |
| <Pause/> with so many insertions and so forth would go away if they were better word models <Pause/> to begin with. |
| </Segment> |
| <Segment StartTime="3130.227" EndTime="3130.574" Participant="me013"> |
| <Pause/> So |
| </Segment> |
| <Segment StartTime="3130.574" EndTime="3132.914" Participant="me013"> |
| <Pause/> this might just be a temporary thing. But - |
| </Segment> |
| <Segment StartTime="3133.890" EndTime="3136.544" Participant="me013"> |
| But, on the <Emphasis> other </Emphasis> hand, and maybe - maybe it's a decent idea. |
| </Segment> |
| <Segment StartTime="3136.544" EndTime="3138.622" Participant="me013"> |
| <Pause/> So um |
| </Segment> |
| <Segment StartTime="3140.100" EndTime="3145.579" Participant="me013"> |
| The question we're gonna wanna go <Pause/> through next week when Hynek shows up I guess is given that we've been - |
| </Segment> |
| <Segment StartTime="3145.579" EndTime="3147.045" Participant="me013"> |
| <Pause/> if you look at what we've been trying, |
| </Segment> |
| <Segment StartTime="3147.045" EndTime="3149.430" Participant="me013"> |
| <Pause/> we're uh looking at |
| </Segment> |
| <Segment StartTime="3149.430" EndTime="3151.790" Participant="me013"> |
| <Pause/> uh, by <Emphasis> then </Emphasis> I guess, |
| </Segment> |
| <Segment StartTime="3151.790" EndTime="3154.091" Participant="me013"> |
| combinations of features and multi-band |
| </Segment> |
| <Segment StartTime="3154.930" EndTime="3160.588" Participant="me013"> |
| Uh, and we've been looking at <Pause/> cross-language, cross <Pause/> task <Pause/> issues. |
| </Segment> |
| <Segment StartTime="3160.588" EndTime="3163.536" Participant="me013"> |
| <Pause/> And they've been not so much looking at |
| </Segment> |
| <Segment StartTime="3162.844" EndTime="3164.382" Participant="me006"> |
| <NonVocalSound Description="mike noise"/> |
| </Segment> |
| <Segment StartTime="3163.536" EndTime="3167.183" Participant="me013"> |
| <Pause/> the cross task uh multiple language issues. |
| </Segment> |
| <Segment StartTime="3166.974" EndTime="3168.067" Participant="me006"> |
| <NonVocalSound Description="mike noise"/> |
| </Segment> |
| <Segment StartTime="3167.183" EndTime="3170.599" Participant="me013"> |
| <Pause/> But they've been looking at uh - <Pause/> at these issues. At the |
| </Segment> |
| <Segment StartTime="3170.599" EndTime="3173.568" Participant="me013"> |
| on-line normalization and the uh |
| </Segment> |
| <Segment StartTime="3173.568" EndTime="3175.357" Participant="me013"> |
| <Pause/> voice activity detection. |
| </Segment> |
| <Segment StartTime="3175.357" EndTime="3178.875" Participant="me013"> |
| <Pause/> And I guess when he comes here we're gonna have to start deciding about |
| </Segment> |
| <Segment StartTime="3178.875" EndTime="3180.422" Participant="me013"> |
| <Pause/> um what do we choose |
| </Segment> |
| <Segment StartTime="3180.422" EndTime="3181.900" Participant="me013"> |
| <Pause/> from what we've looked at |
| </Segment> |
| <Segment StartTime="3181.900" EndTime="3186.556" Participant="me013"> |
| <Pause/> to um blend with <Pause/> some group of things in what they've looked at |
| </Segment> |
| <Segment StartTime="3186.556" EndTime="3188.194" Participant="me013"> |
| <Pause/> And once we <Emphasis> choose </Emphasis> that, |
| </Segment> |
| <Segment StartTime="3188.194" EndTime="3190.225" Participant="me013"> |
| <Pause/> how do we split up the <Pause/> effort? |
| </Segment> |
| <Segment StartTime="3190.930" EndTime="3193.408" Participant="me013"> |
| Uh, because we still have - even once we <Emphasis> choose, </Emphasis> |
| </Segment> |
| <Segment StartTime="3193.408" EndTime="3196.018" Participant="me013"> |
| <Pause/> we've still got <Pause/> uh another |
| </Segment> |
| <Segment StartTime="3196.018" EndTime="3199.052" Participant="me013"> |
| <Pause/> month or so, I mean there's holidays in the way, |
| </Segment> |
| <Segment StartTime="3199.052" EndTime="3200.176" Participant="me013"> |
| but - but uh |
| </Segment> |
| <Segment StartTime="3200.176" EndTime="3204.999" Participant="me013"> |
| <Pause/> I <Emphasis> think </Emphasis> the evaluation data comes January thirty-<Emphasis> first </Emphasis> so there's still a fair amount of time |
| </Segment> |
| <Segment StartTime="3204.999" EndTime="3210.268" Participant="me013"> |
| <Pause/> to <Emphasis> do </Emphasis> things together it's just that they probably should be somewhat more <Emphasis> coherent </Emphasis> between the two <Emphasis> sites </Emphasis> |
| </Segment> |
| <Segment StartTime="3210.268" EndTime="3212.170" Participant="me013"> |
| <Pause/> in that - that amount of time. |
| </Segment> |
| <Segment StartTime="3212.740" EndTime="3220.050" Participant="me018"> |
| When they removed the silence frames, did they insert some kind of a marker so that the recognizer knows it's - <Pause/> knows when it's time to back trace or something? |
| </Segment> |
| <Segment StartTime="3220.050" EndTime="3220.370" Participant="me013"> |
| <VocalSound Description="inbreath"/> |
| </Segment> |
| <Segment StartTime="3220.370" EndTime="3222.662" Participant="me013"> |
| Well, see they, I - I think they're |
| </Segment> |
| <Segment StartTime="3223.720" EndTime="3224.464" Participant="me013"> |
| Um. |
| </Segment> |
| <Segment StartTime="3224.627" EndTime="3225.864" Participant="me013"> |
| I don't know the - |
| </Segment> |
| <Segment StartTime="3225.864" EndTime="3228.837" Participant="me013"> |
| <Pause/> the specifics of how they're doing it. They're - |
| </Segment> |
| <Segment StartTime="3228.837" EndTime="3233.520" Participant="me013"> |
| <Pause/> they're getting around the way the recognizer works because they're not allowed to |
| </Segment> |
| <Segment StartTime="3233.520" EndTime="3236.216" Participant="me013"> |
| <Pause/> um, <Emphasis> change </Emphasis> the <Emphasis> scripts </Emphasis> |
| </Segment> |
| <Segment StartTime="3235.984" EndTime="3237.229" Participant="me018"> |
| Oh, right. |
| </Segment> |
| <Segment StartTime="3236.216" EndTime="3237.867" Participant="me013"> |
| <Pause/> for the <Emphasis> recognizer, </Emphasis> |
| </Segment> |
| <Segment StartTime="3237.867" EndTime="3239.751" Participant="me013"> |
| <Pause/> I believe. So. Uh. |
| </Segment> |
| <Segment StartTime="3238.844" EndTime="3241.350" Participant="me018"> |
| Maybe they're just inserting some nummy frames or something? |
| </Segment> |
| <Segment StartTime="3241.980" EndTime="3243.791" Participant="me013"> |
| Uh, you know <Emphasis> that's </Emphasis> what <Emphasis> I </Emphasis> had thought. |
| </Segment> |
| <Segment StartTime="3244.050" EndTime="3245.549" Participant="me013"> |
| But I don't - I don't think they <Emphasis> are. </Emphasis> |
| </Segment> |
| <Segment StartTime="3245.549" EndTime="3250.848" Participant="me013"> |
| <Pause/> I mean that's - sort of what - the way <Emphasis> I </Emphasis> had imagined would <Emphasis> happen </Emphasis> is that on the other side, yeah you |
| </Segment> |
| <Segment StartTime="3245.551" EndTime="3246.229" Participant="me018"> |
| Hmm. |
| </Segment> |
| <Segment StartTime="3250.848" EndTime="3253.005" Participant="me013"> |
| p- put some low level noise or something. |
| </Segment> |
| <Segment StartTime="3253.500" EndTime="3256.183" Participant="me013"> |
| Probably don't want all zeros. Most recognizers don't like zeros but |
| </Segment> |
| <Segment StartTime="3254.665" EndTime="3255.226" Participant="me018"> |
| Hmm. |
| </Segment> |
| <Segment StartTime="3256.183" EndTime="3257.851" Participant="me013"> |
| <VocalSound Description="laugh"/> but <Pause/> you know, |
| </Segment> |
| <Segment StartTime="3256.243" EndTime="3257.475" Participant="me034"> |
| <VocalSound Description="laugh"/> |
| </Segment> |
| <Segment StartTime="3256.972" EndTime="3257.661" Participant="me018"> |
| Yeah. |
| </Segment> |
| <Segment StartTime="3257.851" EndTime="3261.593" Participant="me013"> |
| <Pause/> put some epsilon in or some rand- sorry epsilon random variable |
| </Segment> |
| <Segment StartTime="3261.593" EndTime="3262.472" Participant="me013"> |
| <Pause/> in or something. |
| </Segment> |
| <Segment StartTime="3262.236" EndTime="3263.770" Participant="me018"> |
| Some constant vector. |
| </Segment> |
| <Segment StartTime="3263.899" EndTime="3268.246" Participant="me013"> |
| <Pause/> Maybe not a constant but it doesn't, uh - don't like to divide by the variance of that, |
| </Segment> |
| <Segment StartTime="3264.442" EndTime="3265.900" Participant="me018"> |
| I mean i- w- |
| </Segment> |
| <Segment StartTime="3266.002" EndTime="3271.063" Participant="me006"> |
| <NonVocalSound Description="mike noise"/> |
| </Segment> |
| <Segment StartTime="3266.345" EndTime="3267.351" Participant="me018"> |
| Or something - |
| </Segment> |
| <Segment StartTime="3268.246" EndTime="3271.950" Participant="me013"> |
| but I mean it's <VocalSound Description="laugh"/> |
| </Segment> |
| <Segment StartTime="3268.936" EndTime="3276.270" Participant="me018"> |
| <VocalSound Description="laugh"/> That's right. But something that - what I mean is something that is <Pause/> very distinguishable from <Pause/> speech. |
| </Segment> |
| <Segment StartTime="3275.108" EndTime="3275.732" Participant="me013"> |
| Mm-hmm. |
| </Segment> |
| <Segment StartTime="3276.270" EndTime="3280.272" Participant="me018"> |
| So that the - the <Emphasis> silence </Emphasis> model in H_T_K will always pick it up. |
| </Segment> |
| <Segment StartTime="3279.648" EndTime="3282.766" Participant="me013"> |
| <Pause/> Yeah. So I - I - that's what I thought they would do. |
| </Segment> |
| <Segment StartTime="3282.766" EndTime="3288.310" Participant="me013"> |
| <Pause/> or else, uh <Pause/> uh maybe there <Emphasis> is </Emphasis> some indicator to tell it to start and stop, I don't know. |
| </Segment> |
| <Segment StartTime="3288.530" EndTime="3289.017" Participant="me018"> |
| Hmm. |
| </Segment> |
| <Segment StartTime="3288.530" EndTime="3292.628" Participant="me013"> |
| But <Emphasis> whatever </Emphasis> they did, I mean they have to play within the rules of this specific evaluation. |
| </Segment> |
| <Segment StartTime="3291.446" EndTime="3292.220" Participant="me018"> |
| Yeah. |
| </Segment> |
| <Segment StartTime="3293.567" EndTime="3295.175" Participant="me013"> |
| <Pause/> We c- we can find out. |
| </Segment> |
| <Segment StartTime="3295.670" EndTime="3299.616" Participant="me018"> |
| Cuz you gotta do <Emphasis> something. </Emphasis> <Emphasis> Otherwise, </Emphasis> if it's just a bunch of <Emphasis> speech, </Emphasis> <Emphasis> stuck </Emphasis> <Emphasis> together </Emphasis> - |
| </Segment> |
| <Segment StartTime="3297.642" EndTime="3299.302" Participant="me006"> |
| <NonVocalSound Description="mike noise"/> |
| </Segment> |
| <Segment StartTime="3299.781" EndTime="3300.400" Participant="me013"> |
| <Emphasis> No </Emphasis> they're - |
| </Segment> |
| <Segment StartTime="3301.320" EndTime="3301.570" Participant="me018"> |
| Yeah. |
| </Segment> |
| <Segment StartTime="3302.410" EndTime="3305.553" Participant="me013"> |
| It would do <Emphasis> badly </Emphasis> and it didn't so <Emphasis> badly, </Emphasis> right? So <Emphasis> they </Emphasis> did <Emphasis> something. </Emphasis> <VocalSound Description="laugh"/> |
| </Segment> |
| <Segment StartTime="3303.277" EndTime="3304.326" Participant="me018"> |
| Yeah, right. <Comment Description="while laughing"/> |
| </Segment> |
| <Segment StartTime="3304.948" EndTime="3305.980" Participant="me018"> |
| Yeah, yeah. |
| </Segment> |
| <Segment StartTime="3306.542" EndTime="3307.220" Participant="me013"> |
| Yeah. |
| </Segment> |
| <Segment StartTime="3307.804" EndTime="3310.962" Participant="me006"> |
| <NonVocalSound Description="mike noise"/> |
| </Segment> |
| <Segment StartTime="3308.543" EndTime="3309.497" Participant="me013"> |
| Uh. |
| </Segment> |
| <Segment StartTime="3310.322" EndTime="3311.830" Participant="me013"> |
| So, O_K, So I think |
| </Segment> |
| <Segment StartTime="3311.830" EndTime="3313.911" Participant="me013"> |
| <Pause/> this brings <Emphasis> me </Emphasis> up to date a bit. |
| </Segment> |
| <Segment StartTime="3313.911" EndTime="3316.200" Participant="me013"> |
| <Pause/> It hopefully brings other <Pause/> people up to date a bit. |
| </Segment> |
| <Segment StartTime="3316.200" EndTime="3317.829" Participant="me013"> |
| <Pause/> And um |
| </Segment> |
| <Segment StartTime="3318.747" EndTime="3320.668" Participant="me013"> |
| Um <Pause/> I think - |
| </Segment> |
| <Segment StartTime="3321.190" EndTime="3327.115" Participant="me013"> |
| Uh, I wanna look at these numbers off-line a little bit and think about it and - <Pause/> and talk with everybody uh, <Pause/> outside of this meeting. |
| </Segment> |
| <Segment StartTime="3327.893" EndTime="3329.749" Participant="me013"> |
| Um, but uh |
| </Segment> |
| <Segment StartTime="3331.940" EndTime="3332.999" Participant="me013"> |
| No I mean it sounds like - I mean |
| </Segment> |
| <Segment StartTime="3332.999" EndTime="3335.495" Participant="me013"> |
| <Pause/> there - there - there are the usual number of - of |
| </Segment> |
| <Segment StartTime="3335.495" EndTime="3339.315" Participant="me013"> |
| <Pause/> little - little problems and bugs and so forth but it sounds like they're getting ironed out. |
| </Segment> |
| <Segment StartTime="3339.315" EndTime="3340.096" Participant="me013"> |
| <Pause/> And now we're |
| </Segment> |
| <Segment StartTime="3340.096" EndTime="3343.614" Participant="me013"> |
| <Pause/> seem to be kind of in a position to actually <Pause/> uh, |
| </Segment> |
| <Segment StartTime="3343.614" EndTime="3346.283" Participant="me013"> |
| <Pause/> <Emphasis> look </Emphasis> at stuff and - and - and <Emphasis> compare </Emphasis> things. |
| </Segment> |
| <Segment StartTime="3346.283" EndTime="3347.897" Participant="me013"> |
| <Pause/> So I think that's - that's pretty good. |
| </Segment> |
| <Segment StartTime="3348.930" EndTime="3349.280" Participant="me013"> |
| Um |
| </Segment> |
| <Segment StartTime="3349.280" EndTime="3350.618" Participant="me013"> |
| <Pause/> I don't know what the - |
| </Segment> |
| <Segment StartTime="3350.618" EndTime="3352.537" Participant="me013"> |
| <Pause/> One of the things I wonder about, |
| </Segment> |
| <Segment StartTime="3352.537" EndTime="3355.830" Participant="me013"> |
| <Pause/> coming back to the first results you talked about, is - is |
| </Segment> |
| <Segment StartTime="3355.830" EndTime="3358.985" Participant="me013"> |
| <Pause/> how much, <Pause/> uh <Pause/> things could be helped |
| </Segment> |
| <Segment StartTime="3358.985" EndTime="3361.915" Participant="me013"> |
| <Pause/> by more parameters. |
| </Segment> |
| <Segment StartTime="3361.915" EndTime="3363.981" Participant="me013"> |
| <Pause/> And uh - |
| </Segment> |
| <Segment StartTime="3363.981" EndTime="3368.489" Participant="me013"> |
| <Pause/> And uh how many more parameters we can afford to <Emphasis> have, </Emphasis> <VocalSound Description="breath-laugh"/> |
| </Segment> |
| <Segment StartTime="3366.144" EndTime="3373.665" Participant="me006"> |
| <NonVocalSound Description="mike noise"/> |
| </Segment> |
| <Segment StartTime="3368.489" EndTime="3370.838" Participant="me013"> |
| <Pause/> in terms of the uh computational limits. |
| </Segment> |
| <Segment StartTime="3370.838" EndTime="3373.350" Participant="me013"> |
| <Pause/> Because <Uncertain> anyway </Uncertain> when we go to |
| </Segment> |
| <Segment StartTime="3373.350" EndTime="3375.513" Participant="me013"> |
| <Pause/> <Emphasis> twice </Emphasis> as much <Emphasis> data </Emphasis> |
| </Segment> |
| <Segment StartTime="3375.513" EndTime="3378.320" Participant="me013"> |
| <Pause/> and have the <Emphasis> same </Emphasis> number of <Emphasis> parameters, </Emphasis> |
| </Segment> |
| <Segment StartTime="3378.320" EndTime="3381.846" Participant="me013"> |
| <Pause/> particularly when it's twice as much data <Emphasis> and </Emphasis> it's quite <Emphasis> diverse, </Emphasis> |
| </Segment> |
| <Segment StartTime="3381.846" EndTime="3385.415" Participant="me013"> |
| <Pause/> um, I wonder if having twice as many parameters would <Emphasis> help. </Emphasis> |
| </Segment> |
| <Segment StartTime="3385.415" EndTime="3389.121" Participant="me013"> |
| <Pause/> Uh, just have a bigger hidden <Emphasis> layer. </Emphasis> |
| </Segment> |
| <Segment StartTime="3386.200" EndTime="3387.030" Participant="mn007"> |
| Mm-hmm. |
| </Segment> |
| <Segment StartTime="3386.782" EndTime="3387.706" Participant="me006"> |
| <NonVocalSound Description="mike noise"/> |
| </Segment> |
| <Segment StartTime="3387.118" EndTime="3387.604" Participant="me034"> |
| <VocalSound Description="breath-laugh"/> |
| </Segment> |
| <Segment StartTime="3390.804" EndTime="3391.470" Participant="me013"> |
| Uh |
| </Segment> |
| <Segment StartTime="3390.871" EndTime="3392.491" Participant="me006"> |
| <NonVocalSound Description="mike noise"/> |
| </Segment> |
| <Segment StartTime="3392.430" EndTime="3392.990" Participant="me013"> |
| But - |
| </Segment> |
| <Segment StartTime="3392.990" EndTime="3393.270" Participant="me034"> |
| <VocalSound Description="inbreath"/> |
| </Segment> |
| <Segment StartTime="3395.342" EndTime="3397.919" Participant="me013"> |
| I <Emphasis> doubt </Emphasis> it would <Pause/> help by forty per <Emphasis> cent. </Emphasis> |
| </Segment> |
| <Segment StartTime="3398.170" EndTime="3400.183" Participant="me013"> |
| But <VocalSound Description="laugh"/> |
| </Segment> |
| <Segment StartTime="3398.701" EndTime="3399.370" Participant="mn007"> |
| <VocalSound Description="laugh"/> |
| </Segment> |
| <Segment StartTime="3399.370" EndTime="3400.113" Participant="mn007"> |
| Yeah. |
| </Segment> |
| <Segment StartTime="3400.183" EndTime="3401.599" Participant="me013"> |
| <Pause/> but uh |
| </Segment> |
| <Segment StartTime="3402.802" EndTime="3403.708" Participant="me013"> |
| <Uncertain> Just </Uncertain> curious. |
| </Segment> |
| <Segment StartTime="3403.708" EndTime="3405.840" Participant="me013"> |
| <Pause/> How are we doing on the |
| </Segment> |
| <Segment StartTime="3405.840" EndTime="3408.732" Participant="me013"> |
| <Pause/> resources? Disk, and - |
| </Segment> |
| <Segment StartTime="3409.986" EndTime="3414.589" Participant="mn007"> |
| I think we're alright, um, <Pause/> not much problems with that. |
| </Segment> |
| <Segment StartTime="3411.787" EndTime="3412.260" Participant="me013"> |
| O_K. |
| </Segment> |
| <Segment StartTime="3414.589" EndTime="3415.384" Participant="mn007"> |
| <VocalSound Description="inbreath"/> |
| </Segment> |
| <Segment StartTime="3415.760" EndTime="3416.666" Participant="me013"> |
| Computation? |
| </Segment> |
| <Segment StartTime="3417.610" EndTime="3418.395" Participant="mn007"> |
| It's O_K. <VocalSound Description="laugh"/> |
| </Segment> |
| <Segment StartTime="3418.145" EndTime="3419.031" Participant="fn002"> |
| <VocalSound Description="laugh"/> |
| </Segment> |
| <Segment StartTime="3418.395" EndTime="3422.521" Participant="mn007"> |
| <Pause/> Well this table took uh <Pause/> more than five days to <Emphasis> get </Emphasis> <Uncertain> back </Uncertain>. |
| </Segment> |
| <Segment StartTime="3418.523" EndTime="3419.448" Participant="me013"> |
| <VocalSound Description="inbreath"/> We - |
| </Segment> |
| <Segment StartTime="3420.823" EndTime="3421.287" Participant="me013"> |
| <VocalSound Description="laugh"/> |
| </Segment> |
| <Segment StartTime="3422.005" EndTime="3423.811" Participant="me013"> |
| Yeah. Yeah, well. |
| </Segment> |
| <Segment StartTime="3422.509" EndTime="3423.616" Participant="me034"> |
| <VocalSound Description="laugh"/> |
| </Segment> |
| <Segment StartTime="3422.770" EndTime="3423.082" Participant="mn007"> |
| <VocalSound Description="laugh"/> |
| </Segment> |
| <Segment StartTime="3424.057" EndTime="3425.141" Participant="mn007"> |
| But - <Pause/> Yeah. |
| </Segment> |
| <Segment StartTime="3425.141" EndTime="3425.965" Participant="mn007"> |
| <VocalSound Description="laugh"/> |
| </Segment> |
| <Segment StartTime="3426.167" EndTime="3430.130" Participant="me013"> |
| Are - were you folks using Gin? <Comment Description="machine name"/> That's a - that just died, you know? |
| </Segment> |
| <Segment StartTime="3430.699" EndTime="3433.580" Participant="mn007"> |
| Mmm, no. You were using Gin <Comment Description="machine name"/> perhaps, yeah? <Pause/> No. |
| </Segment> |
| <Segment StartTime="3433.382" EndTime="3434.774" Participant="fn002"> |
| No. |
| </Segment> |
| <Segment StartTime="3433.989" EndTime="3438.361" Participant="me006"> |
| <VocalSound Description="laugh"/> It just died. <Comment Description="while laughing"/> |
| </Segment> |
| <Segment StartTime="3434.021" EndTime="3436.373" Participant="me013"> |
| No? Oh, that's good. O_K. <Pause/> |
| </Segment> |
| <Segment StartTime="3435.677" EndTime="3437.050" Participant="fn002"> |
| <VocalSound Description="laugh"/> |
| </Segment> |
| <Segment StartTime="3436.373" EndTime="3436.788" Participant="me013"> |
| Yeah, |
| </Segment> |
| <Segment StartTime="3436.788" EndTime="3438.454" Participant="me013"> |
| <Pause/> we're gonna get a replacement |
| </Segment> |
| <Segment StartTime="3437.050" EndTime="3438.264" Participant="fn002"> |
| Yes. |
| </Segment> |
| <Segment StartTime="3438.454" EndTime="3440.286" Participant="me013"> |
| <Pause/> server that'll be a faster server, |
| </Segment> |
| <Segment StartTime="3440.286" EndTime="3441.615" Participant="me013"> |
| <Pause/> actually. That'll be - It's a |
| </Segment> |
| <Segment StartTime="3440.592" EndTime="3441.209" Participant="mn007"> |
| Hmm. <Comment Description="surprised"/> |
| </Segment> |
| <Segment StartTime="3441.615" EndTime="3444.666" Participant="me013"> |
| <Pause/> seven hundred fifty megahertz uh SUN |
| </Segment> |
| <Segment StartTime="3444.455" EndTime="3445.220" Participant="me034"> |
| Tonic. |
| </Segment> |
| <Segment StartTime="3444.666" EndTime="3445.553" Participant="me013"> |
| <Pause/> uh |
| </Segment> |
| <Segment StartTime="3446.377" EndTime="3449.492" Participant="me013"> |
| <Pause/> But it won't be installed for <Pause/> a little while. |
| </Segment> |
| <Segment StartTime="3446.851" EndTime="3447.657" Participant="me034"> |
| <VocalSound Description="laugh"/> |
| </Segment> |
| <Segment StartTime="3450.172" EndTime="3450.701" Participant="mn007"> |
| Mm-hmm. |
| </Segment> |
| <Segment StartTime="3451.978" EndTime="3452.650" Participant="me013"> |
| <VocalSound Description="inbreath"/> |
| </Segment> |
| <Segment StartTime="3452.650" EndTime="3452.959" Participant="me013"> |
| U- |
| </Segment> |
| <Segment StartTime="3452.650" EndTime="3453.590" Participant="me026"> |
| Do we - |
| </Segment> |
| <Segment StartTime="3453.208" EndTime="3453.590" Participant="me013"> |
| Go ahead. |
| </Segment> |
| <Segment StartTime="3453.590" EndTime="3456.980" Participant="me026"> |
| Do we have that big new I_B_M machine the, I think in th- |
| </Segment> |
| <Segment StartTime="3456.408" EndTime="3456.871" Participant="me013"> |
| <VocalSound Description="laugh"/> |
| </Segment> |
| <Segment StartTime="3458.140" EndTime="3460.814" Participant="me013"> |
| We have the <Pause/> little tiny I_B_M machine |
| </Segment> |
| <Segment StartTime="3458.206" EndTime="3459.738" Participant="me034"> |
| <VocalSound Description="laugh"/> |
| </Segment> |
| <Segment StartTime="3460.814" EndTime="3461.601" Participant="me013"> |
| <VocalSound Description="laugh"/> |
| </Segment> |
| <Segment StartTime="3461.601" EndTime="3464.680" Participant="me013"> |
| <Pause/> that might someday grow up to be a big <Pause/> I_B_M machine. |
| </Segment> |
| <Segment StartTime="3464.680" EndTime="3466.145" Participant="me013"> |
| It's got s- slots for eight, |
| </Segment> |
| <Segment StartTime="3466.145" EndTime="3471.026" Participant="me013"> |
| <Pause/> uh I_B_M was donating <Emphasis> five, </Emphasis> I think we only got <Emphasis> two </Emphasis> so far, |
| </Segment> |
| <Segment StartTime="3467.349" EndTime="3469.536" Participant="me006"> |
| <VocalSound Description="laugh"/> |
| </Segment> |
| <Segment StartTime="3471.026" EndTime="3472.900" Participant="me013"> |
| <Pause/> processors. |
| </Segment> |
| <Segment StartTime="3472.900" EndTime="3477.410" Participant="me013"> |
| <Pause/> We had originally hoped we were getting eight hundred megahertz processors. They ended up being five fifty. |
| </Segment> |
| <Segment StartTime="3477.410" EndTime="3483.690" Participant="me013"> |
| So instead of having eight processors that were eight hundred megahertz, we ended up with two <Pause/> that are five hundred and fifty megahertz. |
| </Segment> |
| <Segment StartTime="3483.690" EndTime="3487.683" Participant="me013"> |
| And more are supposed to come soon and there's only a moderate amount of dat- of memory. So I don't think |
| </Segment> |
| <Segment StartTime="3487.683" EndTime="3492.514" Participant="me013"> |
| <Pause/> anybody has been sufficiently excited by it to |
| </Segment> |
| <Segment StartTime="3492.514" EndTime="3494.163" Participant="me013"> |
| <Pause/> spend much time |
| </Segment> |
| <Segment StartTime="3494.163" EndTime="3497.482" Participant="me013"> |
| <Pause/> uh <Pause/> with it, but uh |
| </Segment> |
| <Segment StartTime="3495.164" EndTime="3495.762" Participant="me006"> |
| <VocalSound Description="laugh"/> |
| </Segment> |
| <Segment StartTime="3497.740" EndTime="3498.434" Participant="me013"> |
| <VocalSound Description="inbreath"/> |
| </Segment> |
| <Segment StartTime="3498.434" EndTime="3499.940" Participant="me013"> |
| Hopefully, <Pause/> they'll get us some more |
| </Segment> |
| <Segment StartTime="3499.940" EndTime="3503.416" Participant="me013"> |
| <Pause/> parts, soon and - |
| </Segment> |
| <Segment StartTime="3504.390" EndTime="3506.285" Participant="me013"> |
| Uh, yeah, I think that'll be - once we get it populated, |
| </Segment> |
| <Segment StartTime="3506.285" EndTime="3509.768" Participant="me013"> |
| <Pause/> that'll be a nice <Emphasis> machine. </Emphasis> I mean we <Emphasis> will </Emphasis> ultimately get eight processors in there. |
| </Segment> |
| <Segment StartTime="3510.528" EndTime="3516.235" Participant="me013"> |
| And uh - and uh a nice amount of memory. Uh so it'll be a pr- pretty fast Linux machine. |
| </Segment> |
| <Segment StartTime="3516.677" EndTime="3521.092" Participant="me026"> |
| And if we can do things on Linux, <Pause/> some of the machines we have going already, like Swede? |
| </Segment> |
| <Segment StartTime="3521.040" EndTime="3521.950" Participant="me013"> |
| Mm-hmm. |
| </Segment> |
| <Segment StartTime="3521.950" EndTime="3522.250" Participant="me026"> |
| Um |
| </Segment> |
| <Segment StartTime="3522.680" EndTime="3524.010" Participant="me026"> |
| It seems pretty fast. |
| </Segment> |
| <Segment StartTime="3524.150" EndTime="3524.764" Participant="me013"> |
| Mm-hmm. |
| </Segment> |
| <Segment StartTime="3524.699" EndTime="3525.150" Participant="me026"> |
| But - |
| </Segment> |
| <Segment StartTime="3525.150" EndTime="3526.739" Participant="me026"> |
| I think Fudge is pretty fast too. |
| </Segment> |
| <Segment StartTime="3527.960" EndTime="3531.535" Participant="me013"> |
| Yeah, I mean you can <Emphasis> check </Emphasis> with uh <Pause/> Dave <Emphasis> Johnson. </Emphasis> I mean, it - it's - |
| </Segment> |
| <Segment StartTime="3531.535" EndTime="3533.328" Participant="me013"> |
| <Pause/> I <Emphasis> think </Emphasis> the machine is just <Emphasis> sitting </Emphasis> there. |
| </Segment> |
| <Segment StartTime="3533.328" EndTime="3535.821" Participant="me013"> |
| And it <Emphasis> does </Emphasis> have two <Emphasis> processors, </Emphasis> you know and - |
| </Segment> |
| <Segment StartTime="3535.821" EndTime="3537.263" Participant="me013"> |
| <Pause/> Somebody could do - |
| </Segment> |
| <Segment StartTime="3537.263" EndTime="3539.285" Participant="me013"> |
| <Pause/> you know, uh, check out |
| </Segment> |
| <Segment StartTime="3539.285" EndTime="3542.137" Participant="me013"> |
| <Pause/> uh the multi-threading <Pause/> libraries. And |
| </Segment> |
| <Segment StartTime="3542.137" EndTime="3543.906" Participant="me013"> |
| <Pause/> I mean i- it's possible that the - |
| </Segment> |
| <Segment StartTime="3543.906" EndTime="3548.246" Participant="me013"> |
| <Pause/> I mean, I guess the <Emphasis> prudent </Emphasis> thing to do would be for somebody to do the work on - |
| </Segment> |
| <Segment StartTime="3548.246" EndTime="3550.686" Participant="me013"> |
| <Pause/> on getting our code running |
| </Segment> |
| <Segment StartTime="3550.686" EndTime="3554.172" Participant="me013"> |
| <Pause/> on that machine with <Emphasis> two </Emphasis> processors <Pause/> even though there <Emphasis> aren't </Emphasis> five or eight. |
| </Segment> |
| <Segment StartTime="3554.172" EndTime="3556.964" Participant="me013"> |
| There's - there's - there's gonna be <Emphasis> debugging </Emphasis> hassles |
| </Segment> |
| <Segment StartTime="3556.964" EndTime="3561.171" Participant="me013"> |
| <Pause/> and then we'd be set for when we <Emphasis> did </Emphasis> have five or eight, to have it really be useful. |
| </Segment> |
| <Segment StartTime="3562.098" EndTime="3562.690" Participant="me013"> |
| But. |
| </Segment> |
| <Segment StartTime="3562.690" EndTime="3565.250" Participant="me013"> |
| <Pause/> Notice how I said somebody and |
| </Segment> |
| <Segment StartTime="3565.141" EndTime="3567.523" Participant="me034"> |
| <VocalSound Description="laugh"/> |
| </Segment> |
| <Segment StartTime="3565.250" EndTime="3566.476" Participant="me013"> |
| <VocalSound Description="laugh"/> |
| </Segment> |
| <Segment StartTime="3566.476" EndTime="3569.597" Participant="me013"> |
| turned my head your direction. That's one thing you don't get in these recordings. You don't get the - |
| </Segment> |
| <Segment StartTime="3568.935" EndTime="3570.593" Participant="me018"> |
| <VocalSound Description="laugh"/> |
| </Segment> |
| <Segment StartTime="3569.442" EndTime="3570.255" Participant="me026"> |
| <VocalSound Description="laugh"/> |
| </Segment> |
| <Segment StartTime="3569.597" EndTime="3571.249" Participant="me013"> |
| <Pause/> don't get the visuals but - |
| </Segment> |
| <Segment StartTime="3571.887" EndTime="3580.636" Participant="me026"> |
| I- is it um <Pause/> mostly um the neural network trainings that are <Pause/> um slowing us down or the H_T_K runs that are slowing us down? |
| </Segment> |
| <Segment StartTime="3571.948" EndTime="3574.419" Participant="me006"> |
| <VocalSound Description="laugh"/> <NonVocalSound Description="mike noise"/> |
| </Segment> |
| <Segment StartTime="3572.475" EndTime="3574.182" Participant="me034"> |
| <VocalSound Description="laugh"/> |
| </Segment> |
| <Segment StartTime="3581.150" EndTime="3582.485" Participant="me013"> |
| Uh, I think yes. |
| </Segment> |
| <Segment StartTime="3582.485" EndTime="3584.640" Participant="me013"> |
| <Pause/> Uh, <VocalSound Description="laugh"/> |
| </Segment> |
| <Segment StartTime="3582.948" EndTime="3583.959" Participant="me018"> |
| <VocalSound Description="laugh"/> |
| </Segment> |
| <Segment StartTime="3583.102" EndTime="3584.640" Participant="me034"> |
| <VocalSound Description="laugh"/> |
| </Segment> |
| <Segment StartTime="3585.024" EndTime="3587.823" Participant="me013"> |
| Isn't that right? I mean I think you're - you're sort of held up by both, right? |
| </Segment> |
| <Segment StartTime="3587.823" EndTime="3590.722" Participant="me013"> |
| <Pause/> If the - if the neural net trainings were a hundred times faster |
| </Segment> |
| <Segment StartTime="3590.722" EndTime="3592.969" Participant="me013"> |
| <Pause/> you still wouldn't <Pause/> be anything - |
| </Segment> |
| <Segment StartTime="3592.969" EndTime="3595.094" Participant="me013"> |
| <Pause/> running through these a hundred times faster because you'd |
| </Segment> |
| <Segment StartTime="3595.094" EndTime="3597.480" Participant="me013"> |
| <Pause/> be stuck by the H_T_K trainings, right? |
| </Segment> |
| <Segment StartTime="3595.386" EndTime="3596.066" Participant="mn007"> |
| Mmm. |
| </Segment> |
| <Segment StartTime="3597.084" EndTime="3597.400" Participant="mn007"> |
| Yeah. |
| </Segment> |
| <Segment StartTime="3597.480" EndTime="3599.940" Participant="me013"> |
| <Pause/> But if the H_T_K - I mean I think they're both - |
| </Segment> |
| <Segment StartTime="3599.940" EndTime="3601.330" Participant="me013"> |
| It sounded like they were roughly equal? |
| </Segment> |
| <Segment StartTime="3601.590" EndTime="3602.262" Participant="me013"> |
| Is that about right? |
| </Segment> |
| <Segment StartTime="3602.785" EndTime="3603.317" Participant="mn007"> |
| Yeah. |
| </Segment> |
| <Segment StartTime="3603.526" EndTime="3603.988" Participant="me013"> |
| Yeah. |
| </Segment> |
| <Segment StartTime="3605.207" EndTime="3611.980" Participant="me026"> |
| Because, um <Pause/> I <Emphasis> think </Emphasis> that'll be running Linux, and Sw- Swede and Fudge are already running Linux so, <Pause/> um I could try to |
| </Segment> |
| <Segment StartTime="3612.250" EndTime="3619.475" Participant="me026"> |
| get <Pause/> um the train- the neural network trainings or the H_T_K stuff running under Linux, and to start with I'm <Pause/> wondering which one I should pick first. |
| </Segment> |
| <Segment StartTime="3619.544" EndTime="3620.142" Participant="me013"> |
| <VocalSound Description="inbreath"/> |
| </Segment> |
| <Segment StartTime="3620.142" EndTime="3622.724" Participant="me013"> |
| Uh, probably the neural net cuz it's probably - it - it's - |
| </Segment> |
| <Segment StartTime="3622.724" EndTime="3624.250" Participant="me013"> |
| <Pause/> it's um - |
| </Segment> |
| <Segment StartTime="3624.250" EndTime="3625.470" Participant="me013"> |
| <Pause/> Well, I - I don't know. |
| </Segment> |
| <Segment StartTime="3625.470" EndTime="3626.359" Participant="me013"> |
| <Pause/> They both - |
| </Segment> |
| <Segment StartTime="3626.359" EndTime="3629.747" Participant="me013"> |
| <Pause/> H_T_K we use for |
| </Segment> |
| <Segment StartTime="3629.747" EndTime="3630.535" Participant="me013"> |
| <Pause/> um |
| </Segment> |
| <Segment StartTime="3630.535" EndTime="3632.187" Participant="me013"> |
| <Pause/> this Aurora stuff |
| </Segment> |
| <Segment StartTime="3632.187" EndTime="3633.044" Participant="me013"> |
| <Pause/> Um |
| </Segment> |
| <Segment StartTime="3633.696" EndTime="3635.190" Participant="me013"> |
| <Pause/> Um, I think |
| </Segment> |
| <Segment StartTime="3635.190" EndTime="3637.670" Participant="me013"> |
| <Pause/> It's not <Emphasis> clear </Emphasis> yet what we're gonna use |
| </Segment> |
| <Segment StartTime="3637.670" EndTime="3640.267" Participant="me013"> |
| <Pause/> for trainings uh - |
| </Segment> |
| <Segment StartTime="3640.267" EndTime="3642.404" Participant="me013"> |
| <Pause/> Well, |
| </Segment> |
| <Segment StartTime="3642.404" EndTime="3645.494" Participant="me013"> |
| <Pause/> there's the trainings uh - is it the <Emphasis> training </Emphasis> that takes the time, or the <Emphasis> decoding? </Emphasis> |
| </Segment> |
| <Segment StartTime="3645.494" EndTime="3648.051" Participant="me013"> |
| <Pause/> Uh, is it about equal <Pause/> between the two? |
| </Segment> |
| <Segment StartTime="3648.051" EndTime="3649.830" Participant="me013"> |
| <Pause/> For - for Aurora? |
| </Segment> |
| <Segment StartTime="3649.130" EndTime="3649.994" Participant="mn007"> |
| For H_T_K? |
| </Segment> |
| <Segment StartTime="3650.445" EndTime="3651.962" Participant="me013"> |
| <Pause/> For - Yeah. For the Aurora? |
| </Segment> |
| <Segment StartTime="3652.720" EndTime="3653.320" Participant="mn007"> |
| Uh |
| </Segment> |
| <Segment StartTime="3653.670" EndTime="3655.369" Participant="mn007"> |
| <Emphasis> Training </Emphasis> is longer. <Pause/> |
| </Segment> |
| <Segment StartTime="3655.108" EndTime="3655.630" Participant="me013"> |
| O_K. |
| </Segment> |
| <Segment StartTime="3655.369" EndTime="3655.865" Participant="mn007"> |
| Yeah. |
| </Segment> |
| <Segment StartTime="3657.240" EndTime="3657.855" Participant="me013"> |
| O_K. |
| </Segment> |
| <Segment StartTime="3658.880" EndTime="3660.475" Participant="me013"> |
| Well, I don't know how we can - |
| </Segment> |
| <Segment StartTime="3660.475" EndTime="3661.997" Participant="me013"> |
| <Pause/> I don't know how to - |
| </Segment> |
| <Segment StartTime="3661.997" EndTime="3664.294" Participant="me013"> |
| <Pause/> Do we have H_T_K source? Is that - |
| </Segment> |
| <Segment StartTime="3664.716" EndTime="3665.590" Participant="mn007"> |
| Mmm. |
| </Segment> |
| <Segment StartTime="3665.451" EndTime="3666.010" Participant="me013"> |
| Yeah. |
| </Segment> |
| <Segment StartTime="3668.080" EndTime="3669.975" Participant="me013"> |
| You would <Emphasis> think </Emphasis> that would fairly trivially - |
| </Segment> |
| <Segment StartTime="3669.975" EndTime="3673.038" Participant="me013"> |
| <Pause/> the <Emphasis> training </Emphasis> would, <Emphasis> anyway, </Emphasis> th- the <Emphasis> testing </Emphasis> |
| </Segment> |
| <Segment StartTime="3673.038" EndTime="3674.248" Participant="me013"> |
| <Pause/> uh I don't - I don't |
| </Segment> |
| <Segment StartTime="3674.248" EndTime="3675.668" Participant="me013"> |
| <Pause/> think would |
| </Segment> |
| <Segment StartTime="3675.668" EndTime="3677.215" Participant="me013"> |
| <Pause/> parallelize all that well. |
| </Segment> |
| <Segment StartTime="3677.215" EndTime="3678.797" Participant="me013"> |
| <Pause/> But I think <Pause/> that <Pause/> you could |
| </Segment> |
| <Segment StartTime="3678.797" EndTime="3680.750" Participant="me013"> |
| <Pause/> certainly do d- um, |
| </Segment> |
| <Segment StartTime="3680.750" EndTime="3684.231" Participant="me013"> |
| <Pause/> distributed, sort of - <Pause/> Ah, no, it's the - |
| </Segment> |
| <Segment StartTime="3684.231" EndTime="3685.210" Participant="me013"> |
| <Pause/> each individual |
| </Segment> |
| <Segment StartTime="3685.210" EndTime="3686.316" Participant="me013"> |
| <Pause/> sentence |
| </Segment> |
| <Segment StartTime="3686.316" EndTime="3689.134" Participant="me013"> |
| <Pause/> is pretty tricky to parallelize. |
| </Segment> |
| <Segment StartTime="3689.134" EndTime="3691.879" Participant="me013"> |
| <Pause/> But you could split up the sentences in a test set. |
| </Segment> |
| <Segment StartTime="3690.923" EndTime="3695.170" Participant="me018"> |
| They <Emphasis> have </Emphasis> a - they have a thing for <Emphasis> doing </Emphasis> that and th- they have for <Emphasis> awhile, </Emphasis> in H_T_<Emphasis> K. </Emphasis> |
| </Segment> |
| <Segment StartTime="3695.483" EndTime="3695.807" Participant="me013"> |
| Yeah? |
| </Segment> |
| <Segment StartTime="3695.550" EndTime="3697.890" Participant="me018"> |
| And you can parallelize the training. |
| </Segment> |
| <Segment StartTime="3698.450" EndTime="3701.345" Participant="me018"> |
| And run it on several machines and it just basically keeps counts. |
| </Segment> |
| <Segment StartTime="3698.587" EndTime="3698.990" Participant="me013"> |
| Aha! |
| </Segment> |
| <Segment StartTime="3701.345" EndTime="3702.235" Participant="me018"> |
| And there's something - |
| </Segment> |
| <Segment StartTime="3702.235" EndTime="3705.684" Participant="me018"> |
| <Pause/> a final <Pause/> thing that you run and it accumulates all the counts together. |
| </Segment> |
| <Segment StartTime="3705.628" EndTime="3706.403" Participant="me013"> |
| I see. |
| </Segment> |
| <Segment StartTime="3705.664" EndTime="3706.316" Participant="mn007"> |
| Mmm. |
| </Segment> |
| <Segment StartTime="3706.640" EndTime="3710.320" Participant="me018"> |
| I don't what their scripts are <Pause/> set up to do for the <Emphasis> Aurora </Emphasis> stuff, but - |
| </Segment> |
| <Segment StartTime="3709.838" EndTime="3710.453" Participant="mn007"> |
| Yeah. |
| </Segment> |
| <Segment StartTime="3710.940" EndTime="3713.535" Participant="me013"> |
| Something that we haven't really settled on yet is other than |
| </Segment> |
| <Segment StartTime="3713.535" EndTime="3717.009" Participant="me013"> |
| <Pause/> this <Emphasis> Aurora </Emphasis> stuff, <Pause/> uh what do we do, large vocabulary |
| </Segment> |
| <Segment StartTime="3717.009" EndTime="3719.233" Participant="me013"> |
| <Pause/> training slash testing |
| </Segment> |
| <Segment StartTime="3719.233" EndTime="3721.930" Participant="me013"> |
| <Pause/> for uh tandem systems. |
| </Segment> |
| <Segment StartTime="3721.930" EndTime="3724.670" Participant="me013"> |
| <Pause/> Cuz we hadn't really <Emphasis> done </Emphasis> much with tandem systems for larger stuff. |
| </Segment> |
| <Segment StartTime="3724.670" EndTime="3727.669" Participant="me013"> |
| <Pause/> Cuz we had this one collaboration with C_M_U and we used SPHINX. |
| </Segment> |
| <Segment StartTime="3727.669" EndTime="3731.617" Participant="me013"> |
| <Pause/> Uh, we're also gonna be collaborating with S_R_I and we have their - have <Emphasis> theirs. </Emphasis> |
| </Segment> |
| <Segment StartTime="3731.617" EndTime="3733.743" Participant="me013"> |
| <Pause/> Um <Pause/> So |
| </Segment> |
| <Segment StartTime="3733.743" EndTime="3735.351" Participant="me013"> |
| <Pause/> I don't know |
| </Segment> |
| <Segment StartTime="3736.222" EndTime="3736.905" Participant="me013"> |
| Um. |
| </Segment> |
| <Segment StartTime="3738.229" EndTime="3743.095" Participant="me013"> |
| So I - I think the - the advantage of going with the neural net thing is that we're gonna use the neural net trainings, |
| </Segment> |
| <Segment StartTime="3743.219" EndTime="3744.003" Participant="me013"> |
| no matter what, |
| </Segment> |
| <Segment StartTime="3743.884" EndTime="3744.401" Participant="me026"> |
| O_K. |
| </Segment> |
| <Segment StartTime="3744.227" EndTime="3745.504" Participant="me013"> |
| for a lot of the things we're doing, |
| </Segment> |
| <Segment StartTime="3746.080" EndTime="3747.995" Participant="me013"> |
| whereas, w- exactly which |
| </Segment> |
| <Segment StartTime="3748.664" EndTime="3752.436" Participant="me013"> |
| H_M_M - Gaussian-mixture-based H_M_M thing we use is gonna depend |
| </Segment> |
| <Segment StartTime="3753.009" EndTime="3755.519" Participant="me006"> |
| <NonVocalSound Description="mike noise"/> |
| </Segment> |
| <Segment StartTime="3753.258" EndTime="3753.829" Participant="me013"> |
| uh |
| </Segment> |
| <Segment StartTime="3756.290" EndTime="3757.719" Participant="me013"> |
| <VocalSound Description="inbreath"/> |
| </Segment> |
| <Segment StartTime="3757.719" EndTime="3759.133" Participant="me013"> |
| So with that, |
| </Segment> |
| <Segment StartTime="3759.133" EndTime="3761.560" Participant="me013"> |
| maybe we should uh <VocalSound Description="breath"/> |
| </Segment> |
| <Segment StartTime="3760.297" EndTime="3762.483" Participant="me006"> |
| <NonVocalSound Description="mike noise"/> |
| </Segment> |
| <Segment StartTime="3761.560" EndTime="3762.676" Participant="me013"> |
| go to our |
| </Segment> |
| <Segment StartTime="3762.676" EndTime="3764.906" Participant="me013"> |
| <NonVocalSound Description="pages turning"/> |
| </Segment> |
| <Segment StartTime="3764.588" EndTime="3767.342" Participant="me006"> |
| <NonVocalSound Description="mike noise"/> |
| </Segment> |
| <Segment StartTime="3764.906" EndTime="3767.190" Participant="me013"> |
| digit recitation task. <VocalSound Description="laugh"/> |
| </Segment> |
| <Segment StartTime="3765.862" EndTime="3766.509" Participant="me034"> |
| <VocalSound Description="clear throat"/> |
| </Segment> |
| <Segment StartTime="3766.622" EndTime="3768.754" Participant="fn002"> |
| <VocalSound Description="laugh"/> |
| </Segment> |
| <Segment StartTime="3770.905" EndTime="3772.322" Participant="me006"> |
| <NonVocalSound Description="mike noise"/> |
| </Segment> |
| <Segment StartTime="3773.600" EndTime="3776.546" Participant="me013"> |
| And, it's about eleven fifty. |
| </Segment> |
| <Segment StartTime="3779.287" EndTime="3781.510" Participant="me013"> |
| Canned. <Comment Description="leans into mike, whispered"/> <NonVocalSound Description="spoon dings glass"/> |
| </Segment> |
| <Segment StartTime="3781.770" EndTime="3782.590" Participant="me013"> |
| <VocalSound Description="inbreath"/> |
| </Segment> |
| <Segment StartTime="3784.370" EndTime="3784.800" Participant="me013"> |
| <VocalSound Description="outbreath"/> |
| </Segment> |
| <Segment StartTime="3786.660" EndTime="3788.250" Participant="me013"> |
| Uh, I can - I can start over here. |
| </Segment> |
| <Segment StartTime="3788.510" EndTime="3789.520" Participant="me013"> |
| <VocalSound Description="clear throat, inbreath"/> |
| </Segment> |
| <Segment StartTime="3789.950" EndTime="3793.396" Participant="me013" DigitTask="true"> |
| Two zero one one dash two zero three zero <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="3795.480" EndTime="3796.545" Participant="me013" DigitTask="true"> |
| O_ six nine <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="3797.249" EndTime="3799.261" Participant="me013" DigitTask="true"> |
| zero six zero <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="3799.990" EndTime="3801.100" Participant="me013" DigitTask="true"> |
| one four <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="3801.780" EndTime="3802.298" Participant="me013" DigitTask="true"> |
| two <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="3803.056" EndTime="3806.007" Participant="me013" DigitTask="true"> |
| three zero five one zero eight one <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="3806.554" EndTime="3809.485" Participant="me013" DigitTask="true"> |
| four zero zero four seven two two <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="3810.140" EndTime="3811.094" Participant="me013" DigitTask="true"> |
| six one <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="3811.556" EndTime="3812.892" Participant="me013" DigitTask="true"> |
| seven four two <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="3813.472" EndTime="3814.839" Participant="me013" DigitTask="true"> |
| eight seven eight <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="3813.522" EndTime="3814.629" Participant="fn002"> |
| <VocalSound Description="cough"/> |
| </Segment> |
| <Segment StartTime="3815.194" EndTime="3817.239" Participant="me013" DigitTask="true"> |
| nine nine seven five nine. <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="3817.018" EndTime="3820.014" Participant="me006"> |
| <NonVocalSound Description="mike noise"/> |
| </Segment> |
| <Segment StartTime="3817.604" EndTime="3818.195" Participant="me013" DigitTask="true"> |
| O_. <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="3818.897" EndTime="3819.576" Participant="me013" DigitTask="true"> |
| zero. <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="3820.080" EndTime="3820.932" Participant="me013" DigitTask="true"> |
| one zero <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="3821.500" EndTime="3822.644" Participant="me013" DigitTask="true"> |
| three two two <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="3822.953" EndTime="3824.950" Participant="me013" DigitTask="true"> |
| four three zero zero one <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="3825.496" EndTime="3826.748" Participant="me013" DigitTask="true"> |
| five five <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="3827.170" EndTime="3829.979" Participant="me013" DigitTask="true"> |
| six nine two four zero six three <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="3830.550" EndTime="3831.320" Participant="me013" DigitTask="true"> |
| seven <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="3831.940" EndTime="3832.287" Participant="me013" DigitTask="true"> |
| eight <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="3832.864" EndTime="3834.059" Participant="me013" DigitTask="true"> |
| nine zero three <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="3836.570" EndTime="3836.820" Participant="fn002"> |
| <VocalSound Description="inbreath"/> |
| </Segment> |
| <Segment StartTime="3837.130" EndTime="3843.140" Participant="fn002" DigitTask="true"> |
| Uh, transcript number one nine nine one two O_ one zero. <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="3843.590" EndTime="3844.314" Participant="fn002" DigitTask="true"> |
| O_ nine <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="3844.314" EndTime="3846.751" Participant="fn002" DigitTask="true"> |
| zero eight two seven six <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="3847.061" EndTime="3849.422" Participant="fn002" DigitTask="true"> |
| one nine three three four <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="3850.043" EndTime="3851.900" Participant="fn002" DigitTask="true"> |
| two O_ five five. <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="3852.240" EndTime="3853.359" Participant="fn002" DigitTask="true"> |
| three zero <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="3853.684" EndTime="3854.794" Participant="fn002" DigitTask="true"> |
| five one <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="3855.026" EndTime="3856.541" Participant="fn002" DigitTask="true"> |
| six three two <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="3856.740" EndTime="3857.801" Participant="fn002" DigitTask="true"> |
| seven four <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="3857.960" EndTime="3859.722" Participant="fn002" DigitTask="true"> |
| eight nine one O_ <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="3859.830" EndTime="3860.453" Participant="fn002" DigitTask="true"> |
| nine <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="3860.720" EndTime="3861.068" Participant="fn002" DigitTask="true"> |
| O_ <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="3861.740" EndTime="3864.148" Participant="fn002" DigitTask="true"> |
| zero zero zero seven three <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="3864.650" EndTime="3867.380" Participant="fn002" DigitTask="true"> |
| one two three seven seven four three <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="3867.824" EndTime="3868.632" Participant="fn002" DigitTask="true"> |
| two five <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="3869.022" EndTime="3871.524" Participant="fn002" DigitTask="true"> |
| three six one five <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="3871.998" EndTime="3875.429" Participant="fn002" DigitTask="true"> |
| four eight zero six six zero zero <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="3875.864" EndTime="3876.521" Participant="fn002" DigitTask="true"> |
| five <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="3876.630" EndTime="3877.315" Participant="fn002" DigitTask="true"> |
| six <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="3877.411" EndTime="3878.051" Participant="fn002" DigitTask="true"> |
| seven <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="3878.496" EndTime="3882.075" Participant="fn002" DigitTask="true"> |
| nine one nine six nine five one <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="3883.023" EndTime="3886.740" Participant="me018" DigitTask="true"> |
| Transcript two O_ seven one dash two O_ nine O_ <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="3887.267" EndTime="3888.012" Participant="me018" DigitTask="true"> |
| one <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="3888.891" EndTime="3890.934" Participant="me018" DigitTask="true"> |
| two zero eight four six <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="3891.280" EndTime="3892.101" Participant="me018" DigitTask="true"> |
| four one. <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="3892.790" EndTime="3893.910" Participant="me018" DigitTask="true"> |
| five three <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="3894.240" EndTime="3895.655" Participant="me018" DigitTask="true"> |
| six six zero three <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="3895.655" EndTime="3896.737" Participant="me018" DigitTask="true"> |
| seven <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="3897.169" EndTime="3898.373" Participant="me018" DigitTask="true"> |
| eight nine two <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="3898.769" EndTime="3899.720" Participant="me018" DigitTask="true"> |
| nine <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="3899.720" EndTime="3902.170" Participant="me018" DigitTask="true"> |
| O_ O_ five eight one <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="3902.780" EndTime="3905.140" Participant="me018" DigitTask="true"> |
| one one five three five six four <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="3906.150" EndTime="3907.732" Participant="me018" DigitTask="true"> |
| two seven five six O_ <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="3908.650" EndTime="3910.362" Participant="me018" DigitTask="true"> |
| three six five four zero <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="3911.151" EndTime="3912.081" Participant="me018" DigitTask="true"> |
| four five <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="3912.870" EndTime="3913.372" Participant="me018" DigitTask="true"> |
| five <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="3913.720" EndTime="3915.042" Participant="me018" DigitTask="true"> |
| six <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="3915.700" EndTime="3916.373" Participant="me018" DigitTask="true"> |
| seven <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="3917.270" EndTime="3918.811" Participant="me018" DigitTask="true"> |
| nine one nine four <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="3919.640" EndTime="3921.333" Participant="me018" DigitTask="true"> |
| O_ eight six <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="3921.482" EndTime="3924.360" Participant="me018" DigitTask="true"> |
| zero three one zero nine one two <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="3924.930" EndTime="3926.308" Participant="me018" DigitTask="true"> |
| one eight one zero zero <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="3928.390" EndTime="3932.800" Participant="me034" DigitTask="true"> |
| Transcript two zero five one dash two zero seven zero <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="3934.000" EndTime="3934.990" Participant="me034" DigitTask="true"> |
| zero zero <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="3935.940" EndTime="3937.496" Participant="me034" DigitTask="true"> |
| two two zero four <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="3938.300" EndTime="3939.747" Participant="me034" DigitTask="true"> |
| three two two one three <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="3940.730" EndTime="3941.661" Participant="me034" DigitTask="true"> |
| four six nine <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="3942.155" EndTime="3942.850" Participant="me034" DigitTask="true"> |
| five <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="3943.440" EndTime="3944.179" Participant="me034" DigitTask="true"> |
| six <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="3944.604" EndTime="3945.529" Participant="me034" DigitTask="true"> |
| seven <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="3946.315" EndTime="3948.889" Participant="me034" DigitTask="true"> |
| eight O_ eight eight <Pause/> three seven seven <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="3949.730" EndTime="3951.510" Participant="me034" DigitTask="true"> |
| O_ two eight six one O_ five <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="3952.610" EndTime="3953.329" Participant="me034" DigitTask="true"> |
| zero two <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="3954.650" EndTime="3956.640" Participant="me034" DigitTask="true"> |
| one nine two seven five O_ two <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="3957.390" EndTime="3958.308" Participant="me034" DigitTask="true"> |
| two six <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="3958.881" EndTime="3960.501" Participant="me034" DigitTask="true"> |
| three nine nine eight three <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="3961.220" EndTime="3962.077" Participant="me034" DigitTask="true"> |
| four O_ <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="3962.950" EndTime="3963.501" Participant="me034" DigitTask="true"> |
| five <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="3964.185" EndTime="3966.281" Participant="me034" DigitTask="true"> |
| seven two five six one O_ eight <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="3967.550" EndTime="3968.805" Participant="me034" DigitTask="true"> |
| eight two eight four <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="3969.005" EndTime="4005.536" Participant="me006"> |
| <NonVocalSound Description="mike noise obscures [me006]'s speech, but he is audible on channel E"/> |
| </Segment> |
| <Segment StartTime="3969.980" EndTime="3971.384" Participant="me034" DigitTask="true"> |
| nine six four seven four <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="3972.030" EndTime="3972.409" Participant="me034" DigitTask="true"> |
| O_ <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="3972.699" EndTime="3973.374" Participant="me034" DigitTask="true"> |
| zero <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="3975.400" EndTime="3979.061" Participant="me006" CloseMic="false" DigitTask="true"> |
| Transcript one nine seven one dash one nine nine zero <Comment Description="audible on PZM1, channel E but not on [me006]'s close mike"/> <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="3979.775" EndTime="3980.320" Participant="me006" CloseMic="false" DigitTask="true"> |
| nine <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="3980.948" EndTime="3982.218" Participant="me006" CloseMic="false" DigitTask="true"> |
| O_ eight <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="3982.585" EndTime="3983.315" Participant="me006" CloseMic="false" DigitTask="true"> |
| zero <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="3983.925" EndTime="3985.320" Participant="me006" CloseMic="false" DigitTask="true"> |
| one O_ <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="3985.754" EndTime="3988.926" Participant="me006" CloseMic="false" DigitTask="true"> |
| three one one three zero six four <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="3989.519" EndTime="3990.917" Participant="me006" CloseMic="false" DigitTask="true"> |
| four three <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="3991.473" EndTime="3992.840" Participant="me006" CloseMic="false" DigitTask="true"> |
| five seven two <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="3993.805" EndTime="3997.545" Participant="me006" CloseMic="false" DigitTask="true"> |
| six seven one three eight <Pause/> seven <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="3997.545" EndTime="3998.394" Participant="me006" CloseMic="false" DigitTask="true"> |
| eight <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="3999.043" EndTime="3999.888" Participant="me006" CloseMic="false" DigitTask="true"> |
| nine <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="4000.663" EndTime="4001.755" Participant="me006" CloseMic="false" DigitTask="true"> |
| zero one <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="4002.363" EndTime="4003.430" Participant="me006" CloseMic="false" DigitTask="true"> |
| one two four <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="4004.327" EndTime="4005.464" Participant="me006" CloseMic="false" DigitTask="true"> |
| two two <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="4005.536" EndTime="4019.629" Participant="me006"> |
| <NonVocalSound Description="mike noise - many loud spikes, but his digits are audible on channel E"/> |
| </Segment> |
| <Segment StartTime="4006.306" EndTime="4008.343" Participant="me006" CloseMic="false" DigitTask="true"> |
| three five zero seven <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="4008.737" EndTime="4009.550" Participant="me006" CloseMic="false" DigitTask="true"> |
| four <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="4010.049" EndTime="4010.884" Participant="me006" CloseMic="false" DigitTask="true"> |
| five <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="4011.720" EndTime="4013.220" Participant="me006" CloseMic="false" DigitTask="true"> |
| six zero three three <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="4013.220" EndTime="4017.080" Participant="me006" CloseMic="false" DigitTask="true"> |
| eight one six zero three two five <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="4017.590" EndTime="4020.047" Participant="me006" CloseMic="false" DigitTask="true"> |
| mine one eight seven <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="4020.843" EndTime="4025.944" Participant="me006"> |
| <NonVocalSound Description="mike noise"/> |
| </Segment> |
| <Segment StartTime="4022.280" EndTime="4026.172" Participant="mn007" DigitTask="true"> |
| Transcript two zero three one dash two zero five <Pause/> zero <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="4027.400" EndTime="4028.131" Participant="mn007" DigitTask="true"> |
| zero two <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="4028.830" EndTime="4029.643" Participant="mn007" DigitTask="true"> |
| one two O_ <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="4029.305" EndTime="4038.495" Participant="me006"> |
| <NonVocalSound Description="mike noise"/> |
| </Segment> |
| <Segment StartTime="4030.460" EndTime="4031.555" Participant="mn007" DigitTask="true"> |
| two six six <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="4031.990" EndTime="4033.586" Participant="mn007" DigitTask="true"> |
| three two seven three <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="4034.300" EndTime="4036.076" Participant="mn007" DigitTask="true"> |
| four nine seven nine O_ <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="4036.850" EndTime="4037.498" Participant="mn007" DigitTask="true"> |
| five <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="4037.933" EndTime="4039.643" Participant="mn007" DigitTask="true"> |
| six O_ O_ two <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="4040.650" EndTime="4041.745" Participant="mn007" DigitTask="true"> |
| eight one two <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="4042.540" EndTime="4045.210" Participant="mn007" DigitTask="true"> |
| nine four seven nine one six five <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="4046.080" EndTime="4047.720" Participant="mn007" DigitTask="true"> |
| O_ eight three four <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="4048.550" EndTime="4049.880" Participant="mn007" DigitTask="true"> |
| zero five three <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="4050.710" EndTime="4051.226" Participant="mn007" DigitTask="true"> |
| one <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="4051.750" EndTime="4053.301" Participant="mn007" DigitTask="true"> |
| two O_ six O_ seven <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="4054.008" EndTime="4055.070" Participant="mn007" DigitTask="true"> |
| three zero <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="4055.790" EndTime="4056.607" Participant="mn007" DigitTask="true"> |
| five two <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="4057.220" EndTime="4058.868" Participant="mn007" DigitTask="true"> |
| six four eight eight one <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="4059.710" EndTime="4061.540" Participant="mn007" DigitTask="true"> |
| seven eight six seven four <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="4062.320" EndTime="4064.174" Participant="mn007" DigitTask="true"> |
| eight six one nine four <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="4065.120" EndTime="4065.709" Participant="mn007" DigitTask="true"> |
| nine <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="4066.250" EndTime="4066.612" Participant="mn007" DigitTask="true"> |
| O_ <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="4067.260" EndTime="4071.064" Participant="me026" DigitTask="true"> |
| Transcript one nine three one dash one nine five zero <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="4072.380" EndTime="4072.960" Participant="me026" DigitTask="true"> |
| seven <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="4073.470" EndTime="4073.914" Participant="me026" DigitTask="true"> |
| eight <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="4074.440" EndTime="4075.101" Participant="me026" DigitTask="true"> |
| O_ one <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="4075.580" EndTime="4078.178" Participant="me026" DigitTask="true"> |
| zero two six three five three one <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="4079.090" EndTime="4080.716" Participant="me026" DigitTask="true"> |
| one four five seven <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="4081.440" EndTime="4082.925" Participant="me026" DigitTask="true"> |
| two seven zero one <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="4083.520" EndTime="4084.134" Participant="me026" DigitTask="true"> |
| three <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="4084.279" EndTime="4084.843" Participant="me026" DigitTask="true"> |
| four <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="4085.128" EndTime="4085.813" Participant="me026" DigitTask="true"> |
| five <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="4086.350" EndTime="4088.284" Participant="me026" DigitTask="true"> |
| seven one nine O_ O_ <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="4089.118" EndTime="4090.268" Participant="me026" DigitTask="true"> |
| eight one O_ <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="4090.960" EndTime="4093.561" Participant="me026" DigitTask="true"> |
| nine eight zero seven six zero eight <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="4094.240" EndTime="4095.201" Participant="me026" DigitTask="true"> |
| O_ five one <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="4095.870" EndTime="4096.487" Participant="me026" DigitTask="true"> |
| zero <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="4096.830" EndTime="4097.353" Participant="me026" DigitTask="true"> |
| one <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="4097.650" EndTime="4098.105" Participant="me026" DigitTask="true"> |
| two <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="4098.660" EndTime="4099.470" Participant="me026" DigitTask="true"> |
| four one <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="4100.060" EndTime="4101.934" Participant="me026" DigitTask="true"> |
| five two nine zero four <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="4102.480" EndTime="4103.197" Participant="me026" DigitTask="true"> |
| six <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="4103.860" EndTime="4105.678" Participant="me026" DigitTask="true"> |
| seven eight three six <Comment Description="Digits"/> |
| </Segment> |
| <Segment StartTime="4109.017" EndTime="4112.490" Participant="me013"> |
| Great, uh, could you give Adam a call. Tell him to |
| </Segment> |
| <Segment StartTime="4112.504" EndTime="4117.786" Participant="me006"> |
| Oh. <Comment Description="removes microphone"/> |
| </Segment> |
| <Segment StartTime="4112.820" EndTime="4114.170" Participant="me013"> |
| He's at two nine seven seven. |
| </Segment> |
| <Segment StartTime="4115.220" EndTime="4118.940" Participant="me013"> |
| <Uncertain> O_K. </Uncertain> I think we can <VocalSound Description="inbreath"/> |
| </Segment> |
| <Segment StartTime="4119.340" EndTime="4121.060" Participant="me013"> |
| <Uncertain NumSyllables="1"> @@ </Uncertain> You know Herve's coming tomorrow, right? |
| </Segment> |
| <Segment StartTime="4124.170" EndTime="4127.250" Participant="me013"> |
| Herve will be giving a talk, yeah, talk at eleven. |
| </Segment> |
| <Segment StartTime="4127.033" EndTime="4128.462" Participant="me006" CloseMic="false"> |
| Hello, is Adam there? |
| </Segment> |
| <Segment StartTime="4128.462" EndTime="4129.697" Participant="me006" CloseMic="false"> |
| Hey Adam, this is Barry. |
| </Segment> |
| <Segment StartTime="4129.697" EndTime="4131.584" Participant="me006" CloseMic="false"> |
| Yeah we're all done. |
| </Segment> |
| <Segment StartTime="4131.584" EndTime="4134.133" Participant="me006" CloseMic="false"> |
| O_K, thanks. Bye bye. <NonVocalSound Description="phone hangs up"/> |
| </Segment> |
| <Segment StartTime="4134.287" EndTime="4136.430" Participant="me013"> |
| Did uh, did everybody sign these consent |
| </Segment> |
| <Segment StartTime="4136.970" EndTime="4137.990" Participant="me013"> |
| Er everybody |
| </Segment> |
| <Segment StartTime="4138.802" EndTime="4143.330" Participant="me013"> |
| Has everyone signed a consent form before, on previous meetings? You don't have to do it again each time |
| </Segment> |
| <Segment StartTime="4145.070" EndTime="4145.950" Participant="me013"> |
| Yes. |
| </Segment> |
| <Segment StartTime="4147.040" EndTime="4148.250" Participant="me013"> |
| <Uncertain> microphones off </Uncertain> |
| </Segment> |
|
|
| </Transcript> |
|
|
| </Meeting> |
|
|
|
|
