Requirements
LLMVoice runs locally and expects a working Python audio environment:
- Windows 10/11, 64-bit
- Python 3.11–3.14
- FFmpeg shared build and FFprobe available on your
PATH - PyTorch with a build appropriate for your device
- Enough local storage for XTTS-v2 model weights and generated audio
Install LLMVoice
Clone the LLMVoice repository, then create and activate a virtual environment inside it.
git clone https://github.com/yud3-stack/LLMVoice.git
cd LLMVoice
py -3.11 -m venv .venv
.venv\Scripts\activate
python -m pip install --upgrade pip
Install the shared FFmpeg build required by TorchCodec, reopen the terminal, then confirm both executables are visible:
winget install --id Gyan.FFmpeg.Shared
ffmpeg -version
ffprobe -version
PyTorch and CUDA
Install PyTorch using the command recommended for your operating system and CUDA version. A CUDA-capable NVIDIA GPU makes long-form generation substantially faster. Then install LLMVoice with the XTTS extra:
python -m pip install -e ".[tts]"
CPU generation remains available. Set the device explicitly if auto-detection is not appropriate:
llmvoice config set device cpu
Validate the setup
Run the built-in diagnostic before adding a voice:
llmvoice doctor
Doctor checks Python, FFmpeg, PyTorch, CUDA availability, model access, and writable output locations. Resolve any failed check before a long generation run.