A Local Voice Cloner Dressed Up as OpenAI’s TTS API

What it does This project wraps the open-source Chatterbox TTS engine in a FastAPI server that mimics OpenAI’s TTS endpoints. Existing tools like Open WebUI or AnythingLLM can generate cloned, multilingual speech locally without rewriting their integrations. It also bundles voice library management, automatic text chunking, and an optional React frontend.

The interesting bit The real value is the adapter layer: instead of exposing Chatterbox’s native interface, it translates the familiar OpenAI audio schema so self-hosters can unplug from OpenAI’s API without touching client code. Named voice libraries and per-request parameter tweaks are gravy on top.

Key highlights

• Drop-in OpenAI TTS API compatibility for local stacks
• Voice cloning with uploadable, reusable named voice libraries
• 22 languages with language-aware cloning
• Optional React web UI and FastAPI auto-documentation
• Docker variants for CPU, GPU, and Blackwell (50XX) cards
• Memory monitoring and automatic cleanup

Caveats

• Upstream chatterbox-tts is currently broken on non-CUDA setups, so CPU-only users are stuck until that resolves
• Multilingual support has known issues; the maintainer provides a stable branch if you only need English
• Chatterbox Turbo support is listed as “coming soon” but not yet available

Verdict Worth a look if you already run an OpenAI-compatible local stack and want private voice cloning without API bills. Skip it for now if you are CPU-bound or need production-grade multilingual reliability today.