Chatterbox
resemble-ai/chatterboxReviewed by AI
Checked by RepoPicks Radar on
What is Chatterbox?
A family of text-to-speech models from Resemble AI. You type text and it speaks it, either with its own voice or copying the voice in a short recording you provide, in English or in 23 languages. Small models can even run on a normal processor. You install it with Python (`pip install chatterbox-tts`) and run one of the included example scripts or the small web pages that come with the project; a graphics card makes it much faster. There are also free online demo pages made by the authors. The README says the project was retired in August 2026 and is no longer maintained.
Chatterbox is free and open source (MIT) and can replace ElevenLabs, Murf, and Respeecher.
How easy is it to install Chatterbox?
A · download and open · Web
- Windows C
- macOS C
- Linux C
- Web A
On the web it is an A: the authors keep free demo pages on Hugging Face Spaces where you type text and listen, with nothing to install. On Windows, Mac and Linux it is a C: you install Python, run `pip install chatterbox-tts`, then run one of the included scripts or web pages (`example_tts.py`, `gradio_tts_app.py`), and the model files download on first use; the project was developed and tested on Python 3.11 on Linux, and a graphics card is assumed in the examples (`device="cuda"`), with a separate `example_for_mac.py` for Macs.
Chatterbox on GitHub
- Stars
- 27,748on Nov 10, 2026
- Downloads
- —all releases
- Growth
- +181stars in 6 days
Stars over time
Facts
- Runs on
- Windows, macOS, Linux, Web
- You need to install
- Python, Graphics card
- To build from source
- Python, Graphics card
- Disk space
- 950 MB–4.3 GB · estimate
- Accounts or keys
- None
- Cost
- Free · No price: MIT code, models published on Hugging Face and free demo pages. The README says there is "no support, updates, or paid tier" and points to Resemble AI's own paid speech service for anyone who needs to scale.
- License
- MIT (Permissive)
- Can replace
- ElevenLabs, Murf, Respeecher
- Last change on GitHub
- Nov 4, 2026
Chatterbox risk signals (5)
- Attentiondeclared retired by authors
README.md: "As of August 2026, Chatterbox is retired research and is no longer maintained." and "The code stays public as an archive under the MIT license, with no support, updates, or paid tier." The repository is not archived on GitHub and the last commit in the checked snapshot is 2026-11-04, five days before the check, but there has been no release since v0.1.2 on 2025-06-13 and 381 issues are open.
AI review
- Attentionvoice cloning of real people
The usage examples clone a voice from a short sample: `wav = model.generate(text, audio_prompt_path="your_10s_ref_clip.wav")` with the comment "requires a reference clip for voice cloning". Copying someone's voice without permission is restricted by law in some countries. The only note from the authors is the disclaimer "Don't use this model to do bad things."
AI review
- Infooutput audio is watermarked
README.md: "Every audio file generated by Chatterbox includes Resemble AI's Perth (Perceptual Threshold) Watermarker - imperceptible neural watermarks that survive MP3 compression, audio editing, and common manipulations while maintaining nearly 100% detection accuracy." The README also shows the script that reads the watermark back out of a file.
AI review
- Infodependency installed from a git branch
pyproject.toml installs one dependency straight from GitHub with no fixed version: "resemble-perth @ git+https://github.com/resemble-ai/Perth.git@master", so whatever is on that branch at install time is what you get.
AI review
- Infopinned library versions
pyproject.toml pins exact versions such as "torch==2.6.0", "transformers==5.2.0", "diffusers==0.29.0" and "gradio==6.8.0"; the README explains they are pinned "to ensure consistency", which can clash with other Python programs in the same environment.
AI review
Evidence is shown in its original language.
What Radar could not check about Chatterbox
- How the voices really sound, and how well the voice copying works for each of the 23 languages
- Whether it runs on a Mac or on a computer with no graphics card, and how slow it is (the README claims the small Nano model is 3x faster than realtime on 8 processor cores)
- Real disk space used after the model files download on first use
- Which licence the model files on Hugging Face carry, which can differ from the MIT licence of the code
- Whether the free demo pages and the Hugging Face model downloads stay online now that the authors say the project is retired