Hugging Face's logo

Datasets:
ggfox00000
/
dia-AMICorpus-all

Tasks:
Automatic Speech Recognition
Voice Activity Detection
Modalities:
Audio
Languages:
English
Size:
n<1K
Tags:
ami
meeting
speaker-diarization
diarization
asr
rttm
License:
Dataset card Files
xet
 Community
dia-AMICorpus-all / README.md
ggfox00000's picture
ggfox00000
Dataset card (AMI full mirror)
6cd8d4f verified
preview code
|
raw
history blame contribute delete
5.57 kB
metadata 
license: cc-by-4.0
task_categories:
  - automatic-speech-recognition
  - voice-activity-detection
language:
  - en
size_categories:
  - n<1K
pretty_name: AMI Meeting Corpus — full mirror (Mix-Headset + manual annotations)
tags:
  - ami
  - meeting
  - speaker-diarization
  - diarization
  - asr
  - rttm
  - dialogue-acts
  - nxt
  - audio
annotations_creators:
  - expert-generated
source_datasets:
  - extended|ami-meeting-corpus
dataset_info:
  features:
    - name: audio
      dtype: audio
    - name: file_id
      dtype: string
  splits:
    - name: all
      num_examples: 171

AMI Meeting Corpus — Full Mirror (Mix-Headset + manual annotations)

Miroir complet du AMI Meeting Corpus distribué par l'AMI Consortium (Edinburgh). Tous les fichiers sont repris tels quels depuis la distribution upstream, y compris la structure de dossiers.

Contenu

  • 171 meetings (~100 h d'audio) — scénarios scénarisés (ES/IS/TS) et réunions naturelles (EN/IB/IN)
  • Audio Mix-Headset : amicorpus/<meeting>/audio/<meeting>.Mix-Headset.wav — mixdown des micros serre-tête (un par speaker), c'est le flux audio standard pour benchmark diarisation / ASR close-talk
  • Manual annotations v1.6.2 (NXT XML) dans ami_public_manual_1.6.2/ : segments, words, dialogueActs, topics, abstractive/extractive summaries, namedEntities, focus, gestures (hand/head), movement, disfluency, argumentation, participantRoles, participantSummaries, decision, youUsages, ontologies, corpusResources…
  • Langue : anglais (UK/EU — meetings AMI)
  • Licence : CC-BY-4.0 (cf CCBY4.0.txt à la racine)

Structure 

dia-AMICorpus-all/
├── README.md
├── CCBY4.0.txt
├── amiBuild-*.manifest.txt
├── amicorpus/
│   └── <meeting>/audio/<meeting>.Mix-Headset.wav        # 171 WAV
└── ami_public_manual_1.6.2/                              # NXT v1.6.2
    ├── 00README_MANUAL.txt
    ├── LICENCE.txt
    ├── AMI-metadata.xml
    ├── MANIFEST_MANUAL.txt
    ├── manifest_1.7.html
    ├── resource.xml
    ├── abstractive/            # 142 fichiers
    ├── argumentation/          # 3
    ├── configuration/          # 2
    ├── corpusResources/        # 2
    ├── corpusdoc/              # 73
    ├── decision/               # 1
    ├── dialogueActs/           # 695
    ├── disfluency/             # 160
    ├── extractive/             # 274
    ├── focus/                  # 56
    ├── handGesture/            # 61
    ├── headGesture/            # 173
    ├── movement/               # 498
    ├── namedEntities/          # 468
    ├── ontologies/             # 17
    ├── participantRoles/       # 20
    ├── participantSummaries/   # 323
    ├── segments/               # 687
    ├── topics/                 # 139
    ├── words/                  # 687
    └── youUsages/              # 63

Meetings disponibles (171)

  • Scénarios (ES/IS/TS) : ES2002–ES2016, IS1000–IS1009, TS3003–TS3012
  • Naturels (EN) : EN2001–EN2009
  • Instrumental / natural (IB/IN) : IB4001–IB4011, IN1001–IN1016

Format NXT (exemple segments) 

<?xml version="1.0" encoding="ISO-8859-1"?>
<nite:root nite:id="ES2002a.A.segs" xmlns:nite="http://nite.sourceforge.net/">
  <segment nite:id="ES2002a.sync.1"
           transcriber_start="12.34" transcriber_end="14.56"
           participant="A">
    <nite:child href="ES2002a.A.words.xml#id(ES2002a.w.12)..id(ES2002a.w.27)"/>
  </segment>
  ...
</nite:root>

Pour construire un RTTM standard à partir des segments NXT :

from glob import glob
from xml.etree import ElementTree as ET

NS = {"nite": "http://nite.sourceforge.net/"}
for seg_xml in glob("ami_public_manual_1.6.2/segments/*.segments.xml"):
    root = ET.parse(seg_xml).getroot()
    meeting, spk, _ = seg_xml.split("/")[-1].split(".", 2)
    for seg in root.findall("segment"):
        t0 = float(seg.attrib.get("transcriber_start", seg.attrib.get("starttime", 0)))
        t1 = float(seg.attrib.get("transcriber_end",   seg.attrib.get("endtime",   0)))
        if t1 > t0:
            print(f"SPEAKER {meeting} 1 {t0:.3f} {t1-t0:.3f} <NA> <NA> {spk} <NA> <NA>")

Utilisation 

from huggingface_hub import snapshot_download

root = snapshot_download("ggfox00000/dia-AMICorpus-all", repo_type="dataset")
# root/amicorpus/<meeting>/audio/<meeting>.Mix-Headset.wav
# root/ami_public_manual_1.6.2/segments/<meeting>.<spk>.segments.xml
# root/ami_public_manual_1.6.2/words/<meeting>.<spk>.words.xml

Source

Licence

CC-BY-4.0 — voir CCBY4.0.txt. Distribution miroir de la distribution officielle du AMI Consortium pour faciliter l'accès en recherche.

Citation 

@inproceedings{carletta2005ami,
  author    = {Carletta, Jean and Ashby, Simone and Bourban, Sebastien and Flynn, Mike and Guillemot, Mael and Hain, Thomas and Kadlec, Jaroslav and Karaiskos, Vasilis and Kraaij, Wessel and Kronenthal, Melissa and Lathoud, Guillaume and Lincoln, Mike and Lisowska, Agnes and McCowan, Iain and Post, Wilfried and Reidsma, Dennis and Wellner, Pierre},
  title     = {The {AMI} Meeting Corpus: A Pre-Announcement},
  booktitle = {MLMI},
  year      = {2005},
  pages     = {28--39},
}