Check the environment
Open a new terminal after installation and run:
llmvoice doctor
Doctor verifies the installer-managed Python runtime, FFmpeg Shared, PyTorch, the selected CUDA or CPU profile, and XTTS availability.
Add a voice
Use a clean, single-speaker reference. Approximately 6–30 seconds is recommended.
llmvoice voice add friday reference.wav
llmvoice voice info friday
LLMVoice validates the source and prepares a local reference cache without modifying the original file.
Set a default
llmvoice config set default_voice friday
Generate speech
Create a UTF-8 file named transcript.txt, then run:
llmvoice start transcript.txt
The default output is transcript.mp3 in the transcript directory.
First synthesis may download XTTS-v2 model weights and present the CPML model terms for acceptance. After the model is downloaded, synthesis can run locally and offline. Voice references, transcripts, and generated audio are not uploaded for inference.