[Ready] F5-TTS Training Recipe for WenetSpeech4TTS

[yuekaizhang]

Added F5-TTS Icefall style training codes based on https://github.com/SWivid/F5-TTS.

Training logs and results: https://huggingface.co/yuekai/f5-tts-small-wenetspeech4tts-basic

Model	Seed-TTS test_zh WER	Comment
vall-e 150M	4.33	Icefall codebase reproduced, wenetspeech basic 7k hours
f5-small 150M 32 steps	2.42	Icefall codebase reproduced, wenetspeech basic 7k hours
f5-base 300M 32 steps	1.52	F5-TTS pretrained, Emilia zh&en 100k hours

Samples:

seen1_text19.mov

seen2_text20.2.mov

[yfyeung]

Nice work

[JinZr]

ok! I’ll look into it this weekend Best Regards Jin Yuekai Zhang ***@***.***>于2025年1月20日 周一17:56写道：

…

@yuekaizhang <https://github.com/yuekaizhang> requested your review on: #1846 <#1846> [Ready] F5-TTS Training Recipe for WenetSpeech4TTS. — Reply to this email directly, view it on GitHub <#1846 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AOON42DPXKMMFHAKGV2EE332LTB3DAVCNFSM6AAAAABUEJMCKCVHI2DSMVQWIX3LMV45UABCJFZXG5LFIV3GK3TUJZXXI2LGNFRWC5DJN5XDWMJVHE4TMMJQGA3TINA> . You are receiving this because your review was requested.Message ID: ***@***.***>

[JinZr]

hi! i just finished reviewing this pr and it looks great!

i've left a few comments in it, please take a look when you have a moment

thank you!

[JinZr]

i make a few commits to fix the flake8 and black formatting err, and i think this pr is good to go now.