RVC-Boss /GPT-SoVITS

About GPT-SoVITS

Give GPT-SoVITS a 5-second clip and it reads your text in that voice right away. About a minute of audio gets you a closer match. On a Mac, training runs on CPU for now, since Mac GPU training gave poor results.

Best for: Dubbing teams building character voices from short reference clips.

• Clones a voice from a 5-second sample, no training needed
• Trains on about 1 minute of audio for a closer voice match
• Speaks English, Japanese, Korean, Cantonese and Chinese
• Splits vocals from music, slices audio and writes transcripts
• Picks from Fun-ASR-Nano, SenseVoice or classic FunASR for transcription
• Runs in a WebUI or through API scripts

1. 1
   Make a conda environment: the command below then conda activate GPTSoVits

   conda create -n GPTSoVits python=3.10

   Copy
2. 2
   On Linux or macOS run the command below

   bash install.sh --device <CU126|CU128|ROCM|CPU> --source <HF|HF-Mirror|ModelScope> [--download-uvr5]

   Copy
3. 3
   On Windows run the command below

   pwsh -F install.ps1 --Device <CU126|CU128|CPU> --Source <HF|HF-Mirror|ModelScope> [--DownloadUVR5]

   Copy
4. 4
   If install.sh worked, skip the model download; otherwise put pretrained models in GPT_SoVITS/pretrained_models
5. 5
   Start the WebUI with python webui.py, or double-click go-webui.bat on Windows
6. 6
   For inference only, run the command below

   python GPT_SoVITS/inference_webui.py

   Copy

• GPT-SoVITS vs index-tts