Thanks! And plans to release finetuning code?
Thanks a lot for releasing Avadec! I've been testing it with CSM and the improvement over the default Mimi decoder is really noticeable!
Is there any way to support your work (Ko-fi, donations, etc.)?
Also, would you be willing to share the training / finetuning code for the decoder at some point? I'd love to learn more about how you trained it.
Thanks a lot for releasing Avadec! I've been testing it with CSM and the improvement over the default Mimi decoder is really noticeable!
Is there any way to support your work (Ko-fi, donations, etc.)?
Also, would you be willing to share the training / finetuning code for the decoder at some point? I'd love to learn more about how you trained it.
Hi, I'm glad you liked it!
The training setup is very close to how vocoders and RVQ codecs are trained; you can take a look at DAC repo for reference. it's the standard adversarial training coupled with mel, semantic and stft losses.
due to the situation in Iran, I have a hard time accessing internet and my servers right now, so I can't easily drop the code. but if it's just for learning, taking a look at DAC or xcodec2's repos should suffice.
as for the support I appreciate your kindness; right now the best way to support me is using my work and create cool stuff for the community. you can also have my twitter account in your back pocket since i occasionally announce and explain what i do over there!
if there's anything else i can help with feel free to tell me, i may not be able to respond as quickly as today due to my circumstances, so i apologize in advance.