Summary
Speech you hear, not a file you fetch.
The call ends with the audio already playing, and the progress notifications mean the first run — the one that loads the model — does not look like a hang.
What it is
A text-to-speech server built on the Chatterbox TTS model. One tool takes the text, generates the audio, writes it to a temporary file and plays it — with expressiveness and guidance adjustable on each call.
What you get
- Speech generated from text and played automatically, in a single call
- Expressiveness and classifier-free guidance set per call rather than fixed at startup
- The model loaded on first use, with progress reported through loading, generation and playback
- Temporary files cleaned up after playback, with a configurable audio directory when you want to keep them
- Model status and the device in use — MPS, CUDA or CPU — readable from the server
Requirements
Python with the MCP package, torch and torchaudio, plus the Chatterbox TTS module installed separately. Playback uses afplay, so the automatic playback path is macOS-only; hardware acceleration is used when MPS or CUDA is available.