search

yl4579 / StarGANv2-VC

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

What's the way to finetune pretrained StarGANv2 and ParallelWaveGAN ? #30

Closed skol101 closed 2 years ago

skol101 commented 2 years ago

Granted, there's documentation to fine tune ParallelWaveGAN on a custom dataset, how to do the same with StarGANv2?

skol101 commented 2 years ago

I see there's pretrained_model where one can put the path to the download model.

In this case should num_domains include total numbers of speaks fpre-trained + new) ?

yl4579 commented 2 years ago

No, the num_domains should be the number of speakers in your new dataset.