NTT123/vietTTS

Feature: UnivNet implementation

lethanhson9901 opened this issue · 2 comments

lethanhson9901 commented

Thanks for your great work!
I want to implement univnet vocoder but your model is written in haiku and jax numpy.
https://github.com/mindslab-ai/univnet

I followed your convert code but got stuck:
image
jax numpy has no function unfold or to or transpose ?

How can I fix that problems ?

NTT123 commented

If I were you, I would use the pytorch implementation instead of reimplementing the model in haiku/jax.
You will need to convert the predicted mel spectrogram to pytorch's tensor for your vocoder model.

An example:

mel = HaikuTacotron(text)
mel = jax.device_get(mel)
mel = torch.from_numpy(mel)
wav = TorchVocoder(mel)
  • 👍1
lethanhson9901 commented

Ah, thank you! All done!