search

modelscope / FunCodec

FunCodec is a research-oriented toolkit for audio quantization and downstream applications, such as text-to-speech synthesis, music generation et.al.
https://funcodec.github.io/
MIT License
370 stars 30 forks source link

Release of CN TTS model #13

Closed lucasjinreal closed 9 months ago

lucasjinreal commented 10 months ago

Looks like LauraTTS have Chinese demo, will consider opensource the pretrained model?

ZhihaoDU commented 10 months ago

Currently, the released LauraTTS model is trained on the LibriTTS corpus. I think it is not able to synthesize Chinese directly. The Chinese synthesis model is under internal development and test. We may release it in the future.

lucasjinreal commented 10 months ago

@ZhihaoDU https://lauragpt.github.io/ where does these Chinese synthesis come from?

ZhihaoDU commented 10 months ago

@ZhihaoDU https://lauragpt.github.io/ where does these Chinese synthesis come from?

Chinese cases in this page are generated by LauraGPT rather than LauraTTS, despite they have a similar architecture but different training data. While current released LauraTTS model is trained on LibriTTS, LauraGPT is trained on Librispeech, LibriTTS, AiShell and Aishell-2 and etc.

lucasjinreal commented 10 months ago

Oh, then, I mean LauraGPT, English only model is useless for me.

deyituo commented 10 months ago

@lucasjinreal laoge, you are 666.

lucasjinreal commented 10 months ago

@deyituo Big iron, don't know what u say.