An XTTS server with minimal requirements compatible with SillyTavern.
Create a new environment with mamba:
mamba create -n xtts git python pytorch pytorch-cuda torchaudio -c conda-forge -c nvidia -c pytorch
mamba activate xtts
Clone the repository and install requirements (Build Tools required on Windows):
git clone https://github.com/zuellni/xtts-server
cd xtts-server
pip install -r requirements.txt
Optionally build DeepSpeed on Windows (CUDA Toolkit required):
git clone https://github.com/microsoft/deepspeed
cd deepspeed
build_win.bat
cd dist
pip install deepspeed-X.X.X-cpXXX-cpXXX-win_amd64.whl
Download XTTS-v2, get some speaker files and start the server:
cd xtts-server
git lfs install
git clone https://huggingface.co/coqui/xtts-v2 -b main --depth 1
python server.py -m xtts-v2 -s <speakers_dir>