Hi, about multi-langual support

[MonolithFoundation]

Would try using WavTokenizer for TTS needs support multilangual, for English it should be OK. but for Chinese or Janpanese, what should be done for better adoption?

thanks for this wonderful work!

[jishengpeng]

Would try using WavTokenizer for TTS needs support multilangual, for English it should be OK. but for Chinese or Janpanese, what should be done for better adoption?

thanks for this wonderful work!

Thank you very much for your interest. Building upon the wavtokenizer checkpoint, further training on an additional 10,000 to 20,000 hours of Chinese and Japanese data would likely result in significantly improved performance.

[MonolithFoundation]

Does the training need words level timestamp? 10000 hours needs many audio data, does not very clean (like Emilla dataset) can fill the constraints?