abstruct

"speech_encoder": "vec768l12". more trainning paramaters please find in ATRI_config.json

sovits,diffusion,kmeans moddels included, take it as you need.

performance

a vocal only demo is in the folder.