Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

support fs2 24k training, aligned mel setting with gan vocoder #61

Closed
wants to merge 2 commits into from

Conversation

qinghua2016
Copy link

  • fixed fs2 24k training and inference issues, fixed hifigan vocoder training
  • aligned mels feature for acoustic model and vocoder, so that people can use the pretrained model. fixed other data preprocess issues.
  • fixed librosa usage issues

@RMSnow
Copy link
Collaborator

RMSnow commented Dec 25, 2023

Please use black to format your code. For example, to format a file named wrong_format.py, you can run:

pip install black
black wrong_format.py

@lmxue
Copy link
Collaborator

lmxue commented Dec 25, 2023

  • fixed fs2 24k training and inference issues, fixed hifigan vocoder training
  • aligned mels feature for acoustic model and vocoder, so that people can use the pretrained model. fixed other data preprocess issues.
  • fixed librosa usage issues

Thanks for your suggestion. We tried to use the pre-trained vocoder for fs2 with the settings of 100-dimensional mel-spectrogram, but achieved poor results compared with the current settings. Therefore, we have decided to retain the current settings for FS2.

@VocodexElysium
Copy link
Collaborator

  • fixed fs2 24k training and inference issues, fixed hifigan vocoder training
  • aligned mels feature for acoustic model and vocoder, so that people can use the pretrained model. fixed other data preprocess issues.
  • fixed librosa usage issues

Thanks for your suggestion. As @lmxue mentioned, we have tried to align the FS2 with the default setting of the vocoder in the early stage but resulted in poor performance. If you have got good results with the aligned setting, please attach the pretrained model as well as some demos in this PR and we will process your PR then.

@jiaqili3 jiaqili3 closed this Jul 12, 2024
@jiaqili3
Copy link
Collaborator

Hi, thanks for your suggeston. We're closing this issue for the reasons mentioned by @lmxue and @VocodexElysium , where we tried to use the default vocoder but resulted in poor performance. If you have got good results with the aligned setting, you're welcome to reopen the PR. Thanks!

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

5 participants