Closed dunky11 closed 2 years ago
Hey there, Thanks! I actually used the one available from the mindslab-ai repo. I found it to work fantastic even far out of distribution.
I'd like to train a 44kHz version someday, but I see little benefit in fine-tuning the current version.
Hello first of all great project, this is by far the best zero-shot TTS I've seen yet. I wonder whether the Univnet Vocoder you used was fine-tuned on your dataset or did you simply take the one from https://github.com/mindslab-ai/univnet without further training? I want to fine-tune the vocoder myself and wonder if I should use the generator weights from this project or from the mindslab repo.
Also is there any way to get your UnivNet discriminator weights?