Chatterbox

• Source: https://github.com/resemble-ai/chatterbox (model)
• Server: https://github.com/devnen/Chatterbox-TTS-Server (the image this runs)
• License: MIT
• Alternatives: Speaches, Piper, Coqui TTS, ElevenLabs

Text-to-speech (TTS) service for generating audio from text, over an OpenAI-compatible HTTP API.

:::note Not the Home Assistant one Home Assistant's voice pipeline uses wyoming-piper instead: HA speaks the Wyoming TCP protocol, not HTTP, so it cannot drive Chatterbox without a bridge.

Reach for Chatterbox when something in the stack is generating audio over HTTP and you want an expressive or cloned voice. :::

:::caution Generated audio is watermarked Chatterbox stamps every waveform it produces with Resemble AI's Perth imperceptible watermark, in the model package itself (chatterbox/tts.py applies PerthImplicitWatermarker on every synthesis). There is no setting to turn it off. :::