BASE MODEL · TTS

GPT-SoVITS

Few-shot voice cloning loved by the anime and character community.

Voices built on GPT-SoVITS

Technical details

Architecture
GPT semantic model + SoVITS vocoder
Type
TTS
Size
0.8 GB
Runs on
~4 GB VRAM
Languages
Chinese, English, Japanese, Korean, Cantonese
Recommended settings
top_k=15 · top_p=1.0 · temperature=1.0
Version
v2 · updated 1 mo ago
Source
GitHub
License
MIT · Commercial OK

GPT-SoVITS is developed and owned by its original authors. timbre only indexes it.

COMPARE Compare