-
Notifications
You must be signed in to change notification settings - Fork 25
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
is pretrained weight of discriminator of base model available? #5
Comments
@bshall Thank you for this great work @bshall @seastar105 only loading generator weights and starting fine-tuning is a good idea?
Thanks |
@mraj96 maybe discriminator will be too weak. but I'm gonna try it. training from scratch takes about 1 week in my environment. It sounds much doable for me. |
@mraj96 finetuning or training hifigan takes really long time, so i gave up. i've finetuned vocoder with only generator, but ~300 epoch trainig showed nothing difference between using base vocoder. using base vocoder works pretty good if you trained with clean dataset, such as kss, jsut. but my own dataset is quite noisy, but audible showed really poor quality. this makes finetuning necessary. so i gave up and trying to use hubert-soft and e2e model(e.g. VITS) together. |
Thanks for nice work. @bshall
I'm trying to train hifigan now, but it takes so long training it from scratch using other dataset.
If discriminator of base model is also available, I could start finetuning based on that vocoder. it seems that you released only generator. Could you also release discriminator weights?
The text was updated successfully, but these errors were encountered: