chatterbox-tts-api

Why we included this project

Self-hosted chat interfaces like Open WebUI and AnythingLLM can already call OpenAI's speech API, but that means shipping audio to a third party and paying per character. This project swaps in a local endpoint that speaks the same protocol, so those tools can generate speech without leaving your machine. It runs the Chatterbox model behind a FastAPI service, supports 22 languages, and can clone a voice from your own samples, which is handy when you want the same voice across several apps. A built-in voice library and automatic chunking of long text make it usable beyond short replies, and the Docker images cover everything from CPU-only hosts to Blackwell GPUs, so deployment mostly comes down to picking a compose file. If you care about privacy or just want to avoid per-character bills, this is a clean way to add TTS to tooling you already run.