Chatterbox
Open Weights
Resemble AI
Resemble AI's MIT-licensed open-source text-to-speech family, offering zero-shot voice cloning from a short reference clip. It ships as Turbo (350M, English, sub-200ms latency), Nano (110M, English, roughly 3x faster than realtime on 8 CPU cores) and Multilingual V3 (500M, 23+ languages), plus dedicated single-language finetunes. Every generated file carries Resemble's Perth perceptual watermark, which survives MP3 compression and common editing.
Strengths
- MIT licensed - among the most permissive licences for a production TTS model
- Zero-shot voice cloning from a short reference clip
- Turbo (350M) targets sub-200ms latency for production use
- Nano (110M) runs roughly 3x faster than realtime on 8 CPU cores
- Multilingual V3 (500M) covers 23+ languages including Arabic, Hindi, Japanese and Swahili
- Built-in Perth watermarking survives MP3 compression and common audio editing
Caveats
- Zero-shot voice cloning carries obvious misuse risk - the watermark mitigates but does not prevent it
- Resemble publishes Podonos preference evaluations without stating the percentages, so no scores are recorded here
- English-only for the Turbo and Nano variants; multilingual needs the larger V3
- No release date is stated on the repository
- Vendor
- Resemble AI
- License
- MIT
- Modalities
- textaudio
Capabilities
Vision
Audio
Video
Tool Use
Resources
Pricing
Open weights under the MIT licence - self-host at no licence cost. Nano runs faster than realtime on 8 CPU cores, so CPU-only deployment is viable.