svc-develop-team/so-vits-svc
🔧 Toolsvc-develop-team
High-fidelity singing voice conversion via SoftVC VITS.
So-VITS-SVC implements a SoftVC VITS (Variational Inference with adversarial learning for Text-to-Speech) model for singing voice conversion. It uses a soft speech encoder (SoftVC) to extract content features, a VITS decoder to generate waveforms, and a flow-based model for high-fidelity reconstruction. The training pipeline includes GAN-based adversarial training and fine-tuning on target speaker data. Key features include multi-speaker support, real-time inference, and pitch control. The repository provides scripts for data preprocessing, training, and inference, along with pretrained models. It has gained popularity for creating AI-generated singing covers and voice morphing.
💡Highlights
- ├─28k+ GitHub stars
- ├─SoftVC + VITS architecture
- └─Real-time voice conversion
🎯For
- ├─Music producers
- ├─AI researchers
- └─Voice actors