yl4579 / StarGANv2-VC

StarGANv2-VC: A Diverse, Unsupervised, Non-parallel Framework for Natural-Sounding Voice Conversion
MIT License
466 stars 110 forks source link

about high samplerate training #53

Closed MMMMichaelzhang closed 1 year ago

MMMMichaelzhang commented 2 years ago

I want to retrain vocoder for higher samplerate,do you have plan to update the vocoder training ?how about the vocoder preprocessing?

Kreevoz commented 2 years ago

Since you now have access to the training code for the ASR and F0 models, you can replace the preprocessing with whatever you like and train a vocoder with that too - you can adjust the preprocessing code of the PWG repo as you see fit. But you'd have to train all models from scratch with your new settings.