Chatterbox
Open-source text-to-speech models that can clone a voice from a short reference clip.
- Stars
- 27.5k
- Forks
- 3.7k
- Licence
- MIT
- Language
- Python
- Last update
- 4 months ago
- Type
- AI model / agent
What it is
Chatterbox is a family of open-source text-to-speech models from Resemble AI. You give it text and, for voice cloning, a short reference audio clip, and it generates spoken audio in that voice. The family includes English-focused models (Turbo, Nano), a multilingual model (V3), and the original Chatterbox.
- Generate English speech from text with a voice cloned from a roughly 10-second reference clip
- Add paralinguistic tags such as [cough] or [laugh] to make speech sound more natural
- Produce multilingual voice-cloned audio with the V3 model
- Run a smaller English model on a CPU for on-device experiments
Great for
- Developers building voice agents who need low-latency English speech
- Teams wanting one open model for voice cloning across many languages
- People running speech generation on CPU or limited hardware (Nano)
- Creators who want expressive narration with tags like [laugh] or [chuckle]
Not ideal for
- People who need a no-code tool: the README only shows Python code examples
- Users who need guaranteed production-grade scaling; the README points to a paid service for tuning and higher accuracy
- Anyone needing a language outside the listed coverage, since the README gives '23+' languages without a full list in the excerpt
How to get it
- 1
Install with a package manager
You'll need: Python 3.11 (the version the authors tested on)
Install from PyPI
pip install chatterbox-ttsClone the repository
git clone https://github.com/resemble-ai/chatterbox.gitInstall from source in editable mode
pip install -e . - 2
Run it
- Install the package with pip install chatterbox-tts
- Load a model, for example ChatterboxTurboTTS.from_pretrained(device="cuda")
- Call model.generate(text, audio_prompt_path="your_10s_ref_clip.wav")
- Save the output with torchaudio.save using model.sr as the sample rate
- 3
Get the most out of it
- Use Turbo for low-latency English voice agents and Nano for CPU or tight memory budgets
- Add paralinguistic tags like [chuckle] inside your text for more realistic delivery
- Use the Multilingual V3 model with t3_model="v3" for cross-language voice cloning
- Check the demo page and model zoo table to compare quality before choosing a model
Downloads
Strengths
- The README reports state-of-the-art quality claims and links to public demo samples
- Permissive MIT license and a large community (about 27k stars)
- Offers models at several sizes, from 110M (Nano) to 500M (Multilingual), to fit different hardware
Trade-offs
- Models need a reference clip for voice cloning, and the usage examples rely on Python and PyTorch with a GPU ('cuda') by default
- Benchmark and quality claims come from the authors and vendor-linked evaluations, not independent tests in the README
- The README is truncated, so full language list, hardware requirements, and license details beyond MIT are unclear
How it compares
| Project | Stars | Type | Level | Updated | Best for |
|---|---|---|---|---|---|
| chatterbox (this) | 27.5k | AI model / agent | intermediate | 4 months ago | Developers building voice agents who need low-latency English speech |
| piper | 11.3k | App | intermediate | 1.2 years ago | Developers who want offline speech output in a local app or home automation setup |
| bark | 39.3k | AI model / agent | intermediate | 2.2 years ago | Developers who want to add text-to-speech or sound generation to a Python project |
Stats synced from GitHub. The assessment above was drafted with AI from the project's README on 2026-10-29 and may miss recent changes — check the README before relying on a command. MIT licence: free to use, modify and distribute, including commercially, as long as you keep the copyright notice.