neonbjb / tortoise-tts

A multi-voice TTS system trained with an emphasis on quality
Apache License 2.0
13.25k stars 1.84k forks source link

get no benefit from the CLVP module #803

Open JohnHerry opened 4 months ago

JohnHerry commented 4 months ago

Hi, all I see no benefit from the CLVP module, the best score AR generated mel code may not so good, even with some timbre mixture, Should we put the speaker cond into the text tokens during training CLVP?

hoyden commented 4 months ago

You can remove CLVP module, it's useless in my test. RLHF is somewhat more useful.

JohnHerry commented 4 months ago

Thank you, and is there any more detailed information about the RLHF you are taking for this GPT generation?