resemble-ai/chatterbox is an open-source text-to-speech (TTS, the technology that turns text into speech) model family released by Resemble AI. It is implemented in Python and distributed under the MIT license. The repository was created on April 23, 2025, and the latest release published on GitHub is v0.1.2.
The models differ by use. Chatterbox Multilingual supports speech generation in multiple languages and voice cloning (reproducing a specific person's vocal characteristics), while Chatterbox-Turbo and Chatterbox-Nano are used for English speech generation. The README states that these two models handle non-linguistic expressions such as laughter as tags. The repository has 27,122 GitHub stars.
However, audio produced by Chatterbox carries a Perth (Perceptual Threshold) watermark. A watermark is a technique that leaves an identifying mark in a speech signal. The available materials contain no release notes for v0.1.2, so what changed in it cannot be known. There are no figures comparing its performance with other speech synthesis models, and no benchmark remains that would back up the level the repository description claims. Among the scored items, only momentum is given, at 5.8 out of 10, while freshness, technology, and quality scores are not provided. Taking the watermark and the lack of performance data together, it is more accurate at present to treat this as a reference resource for speech synthesis whose scope and license are clear.