svc-develop-team/so-vits-svc

Uses a SoftVC content encoder to extract speech features directly from audio, feeding them into VITS while preserving pitch and intonation without text-based intermediate representations. Replaces the vocoder with NSF HiFiGAN to eliminate sound artifacts, and supports multiple encoders including ContentVec and Whisper-PPG. Includes shallow diffusion refinement, feature retrieval integration with RVC, and loudness embedding for enhanced voice conversion quality.

28,008 stars. No commits in the last 6 months.