Chatterbox Multilingual V3
Description
Chatterbox Multilingual V3 is an open text-to-speech model from Resemble AI with 0.5 billion parameters. Per the model card, it speaks 23 languages, clones a voice from a sample and, in V3, has better speaker similarity and fewer hallucinations (unwanted repetition or off-text speech) than the earlier multilingual version. The family also includes the original English model with a control for how strongly emotion is expressed, and six single-language fine-tunes. Every generated audio file carries an imperceptible PerTh watermark built into the model. The authors do not state a VRAM requirement. The weights are public on Hugging Face without any acceptance step; V3 was added on 2026-06-10, the first version came out in May 2025. The license is MIT (per the model card and the GitHub repository), which allows commercial use. The authors offer an official demo Space on Hugging Face and a demo page with audio samples, and the code is on GitHub. Voice cloning should only be used with the speaker's consent. Date of check: 2026-10-19.
Pros and cons
Pros
- 23 languages and voice cloning from a sample, per the model card
- Small model: 0.5 billion parameters
- MIT license, weights open without an acceptance step
- Built-in watermark on every generated file
- Official Space for trying it without installation
Cons
- No voice design from a text description
- The authors do not state VRAM requirements
- The emotion-strength control is described for the English model
- The watermark cannot be turned off in the model's design
Key facts
- License
- MIT
- Commercial use
- Allowed
- Developer
- Resemble AI
- Interface languages
- English
- Checked on
- Oct 19, 2026