ChatterboxUnclaimed

Chatterbox is an open source text-to-speech model from Resemble AI, released under the MIT license. It turns text into speech and can copy a voice from a reference clip of about five seconds, with no training step. It also offers emotion control, which adjusts how expressive the speech sounds. In a blind evaluation run through Podonos, the page says 63.75% of evaluators preferred Chatterbox over ElevenLabs.

The family has three versions. Chatterbox is the original. Chatterbox Multilingual supports 23 languages with the same voice cloning. Chatterbox Turbo is aimed at the fastest inference and adds tags for non-speech sounds such as laughter and breaths.

It is built for developers who want to add speech to their own apps, voice agents or media, and for teams that need to run it on their own hardware. The page says it can be deployed locally, on a GPU or fully on-premise.

To start, install the chatterbox-tts package from pip or clone the repository from GitHub. Point the model at a short reference clip and pass in the text, and it generates the audio. The page says no API keys, rate limits or sign-ups are needed. Only clone voices you have permission to use.

Every generated file carries PerTh, an imperceptible watermark from Resemble AI that marks the audio as created by Chatterbox. The site says removing the watermark goes against the model's intended use.

On cost, the page says Chatterbox is free to use, including commercially, with no royalties or usage caps. It also notes that Chatterbox and Resemble's other generative voice models remain open under Voice AI Research but are no longer offered as commercial paid products. The paid plans on Resemble's pricing page cover its detection products, not Chatterbox.