search

TensorSpeech / TensorFlowTTS

:stuck_out_tongue_closed_eyes: TensorFlowTTS: Real-Time State-of-the-art Speech Synthesis for Tensorflow 2 (supported including English, French, Korean, Chinese, German and Easy to adapt for other languages)
https://tensorspeech.github.io/TensorFlowTTS/
Apache License 2.0
3.8k stars 810 forks source link

Italian models #730

Closed grausof closed 2 years ago

grausof commented 2 years ago

Hello, I'm trying to train the model also for Italian language and I would like to use the MLS dataset which always derives from LibriVox. My question is whether I have to use a single speaker in the dataset, or can I use more than one. Also I ask you if you have already tried this dataset! Thanks

ZDisket commented 2 years ago

@grausof You probably want to try single speaker first. You can try multiple speakers but you'll have to modify the dataloader and Tacotron2 multispeaker seems broken, as it failed to learn attention at all during all my experiments, which I think is the fault of adding speaker embeddings to both encoder and decoder, keith ito's implementation only has them on the encoder.

stale[bot] commented 2 years ago

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs.