Inquiry About the Language Distribution of the 200k-Hour Dataset and FACodec Support for Chinese

[xjf-303]

@lifeiteng Thank you for your excellent work on NaturalSpeech 3! I have a couple of questions regarding the dataset and FACodec:

Regarding the 200k-hour internal dataset mentioned in the paper:

Could you please clarify the language distribution of this dataset? Is it predominantly English, or does it include other languages such as Chinese?

About FACodec:

Is FACodec designed to support Chinese audio, especially for extracting content, prosody, acoustic details, and timbre in Chinese speech?
If not, would additional training or fine-tuning on Chinese data be required to adapt FACodec for Chinese speech synthesis?
Your insights on this would be very helpful! Thank you for your time and for sharing your incredible research.