The vanilla NaturalSpeech2 train on WenetSpeech4TTS using Amphion tooltik.
The entire training process follows its training code, except that the text-to-phoneme feature step and text encoder are slightly different.
Checkpoints
- ns2_base_model.bin : NaturalSpeech2 trained with the WenetSpeech4TTS Basic subset
- ns2_38_model.bin : NaturalSpeech2 Basic fine-tuning with the WenetSpeech4TTS Standard subset
- ns2_4_model.bin : NaturalSpeech2 Standard fine-tuning with the WenetSpeech4TTS Premium subset
usage
coming soon.