Check the environment
Begin with the diagnostic command. It is the fastest way to catch missing audio or compute dependencies.
llmvoice doctor
Add a voice
Register a reusable voice named friday from a clean WAV reference:
llmvoice voice add friday reference.wav
LLMVoice inspects the source file and stores the voice locally. Use a clear recording with one speaker and little background noise.
Set a default
Make the new voice the default for future runs:
llmvoice config set default_voice friday
You can also set a default language:
llmvoice config set default_language en
Generate speech
Pass a UTF-8 text transcript to the start command:
llmvoice start transcript.txt
LLMVoice chunks the text, generates each segment locally with XTTS-v2, merges the audio, and writes transcript.mp3.