facebookresearch/textlesslib

Tacotron2 training codes

Opened this issue · 1 comments

Hi, thank you very much for the great work!

Could you please release the training codes to convert discrete ids to mel-spectrograms?

I was wondering how to train this tacotron2 model. The ground truth mel-spectrograms is sythesized by a TTS model?

Thank you!

Hi,

Please see issue #3951 in fairseq. You can refer to the fairseq unit-to-speech example here, or the NVIDIA's Tacotron2 repo link.