Seed-VC Real-time Voice Conversion separated for easier deployment and usage.

This project provides a standalone implementation of the real-time voice conversion functionality from Seed-VC, supporting zero-shot voice conversion with low latency.

• Real-time Voice Conversion: Capable of cloning voices with 1~30 seconds of reference speech.
• Low Latency: Algorithm delay of ~300ms and device side delay of ~100ms.
• Zero-shot: No training required for new voices.

• OS: Windows or Linux.
• GPU: NVIDIA GPU with CUDA support is strongly recommended for real-time performance.
• Python: 3.10

1. Create and activate a virtual environment:

   # Windows
   python -m venv venv
   venv\Scripts\activate
   
   # Linux
   python3 -m venv venv
   source venv/bin/activate

2. Install dependencies:

   Note: The default requirements.txt includes CUDA 12.1 support. If you have a different CUDA version, please edit requirements.txt accordingly.

   pip install -r requirements.txt

3. Run the GUI:

python real-time-gui.py --checkpoint-path <path-to-checkpoint> --config-path <path-to-config>

• checkpoint is the path to the model checkpoint if you have trained or fine-tuned your own model, leave to blank to auto-download default model from huggingface. (seed-uvit-tat-xlsr-tiny)
• config is the path to the model config if you have trained or fine-tuned your own model, leave to blank to auto-download default config from huggingface

It is strongly recommended to use a GPU for real-time voice conversion. Below are some benchmark results on an NVIDIA RTX 3060 Laptop GPU:

• Diffusion Steps: 4~10 recommended for fastest real-time inference.
• Block Time: The length of each audio chunk. Must be greater than inference time per block.
• Extra context: Increasing this improves stability but adds latency.
• Virtual Cable: Use VB-CABLE to route GUI output to a virtual microphone for use in other applications (Discord, Zoom, etc.).

This project is a separated version of Seed-VC. All credit goes to the original authors.

• Seed-VC - Original Project
• Amphion for providing computational resources and inspiration!
• Vevo for theoretical foundation of V2 model
• MegaTTS3 for multi-condition CFG inference implemented in V2 model
• ASTRAL-quantiztion for the amazing speaker-disentangled speech tokenizer used by V2 model
• RVC for foundationing the real-time voice conversion
• SEED-TTS for the initial idea