search

unilight / s3prl-vc

S3PRL-VC: A Voice Conversion Toolkit based on S3PRL
https://unilight.github.io/Publication-Demos/publications/s3prl-vc/index.html
MIT License
93 stars 12 forks source link

How to generate conf/f0.yaml #5

Closed tan90xx closed 5 months ago

tan90xx commented 5 months ago

Hi, thank you for such an amazing project!

I am unsure how to generate the conf/f0.yaml file. Is there any difference between all setting to defaults (f0min=40, f0max=500) or per-speaker settings?

unilight commented 5 months ago

Hi @tan90xx,

To answer your first question, you can try to use s3prl_vc/bin/create_histogram.py. The code should be self-contained, but if you have problems executing it, please let me know. For your second question, the f0 range is mainly used for better WORLD feature extraction during MCD evaluation, so setting it to default will not hurt the model performance.

Hope this helps!

tan90xx commented 5 months ago

Thank you so much!