search

Tinglok / CVC

CVC: Contrastive Learning for Non-parallel Voice Conversion (INTERSPEECH 2021, in PyTorch)
MIT License
57 stars 12 forks source link

Evaluation of CycleGAN-VC3 in your paper #10

Closed powei-C closed 2 years ago

powei-C commented 2 years ago

Hi, Thank you for sharing the paperwork! I wonder if you trained the Resnet-9blocked generator in CycleGAN-VC3 for your paper evaluation. Since I followed your command instruction for training CyleGAN-VC3, it uses the same "netG" as the CVC framework.

Tinglok commented 2 years ago

Hi @powei-C, Thanks for your interest in our work. Actually we used the Resnet-9blocked generator in CycleGAN-VC3 for fair comparison in our paper, since we'd like to demonstate that cycle-consistency loss can be replaced by the contrastive loss.

powei-C commented 2 years ago

Thank you for your explanation!

I still have one question. When listening to the demo audio, I found that the audio's speed slows down a little bit compared to the source audio, even though I retrain the model and test it, there is a similar situation. Is it reasonable?

Tinglok commented 2 years ago

This may be because the hyperparameters (win_length, hop_size, etc) of PWG vocoder do not match my Mel-spectrogram extraction code here. Therefore you could make them consistent, which may solve the problem. Please keep me posted if it works.

powei-C commented 2 years ago

Thank you for your instant reply! It makes sense!!! Since I have run out of my GPU resources recently, I may test it a few days later.

Best Regards

powei-C commented 2 years ago

It worked! Thank you for your help!

Best Regards

Quetzalquatl commented 1 year ago

It worked! Thank you for your help!

Best Regards

@powei-C How did you do the settings?

powei-C commented 1 year ago

@Quetzalquatl It's been a long time... I am not 100% sure right now. I remember that I just adjust the parameter's value in CVC/data/wav_folder.py to match the one in "/CVC/checkpoints/vocoder/config.yml" such as hop size, fft size and win_length .... Hope this help~

Quetzalquatl commented 1 year ago

@powei-C I changed the settings in wav_folder.py according to the parameters in vocoder/config.yml but the result is very bad. no matter what i did i couldn't fix it well. If yes, could you paste your settings here?