Chatterbox TTS
Retired open-source text-to-speech model family with zero-shot voice cloning, kept public as an archive.
MIT code for Turbo, Nano and Multilingual V3 speech models, with no support since August 2026.
Chatterbox is a family of open-source text-to-speech models from Resemble AI that clone a voice from a short reference clip. The README says that as of August 2026 it is retired research and no longer maintained. Use it only if you accept an archive with no support or updates.
What it is
You install the chatterbox-tts Python package and generate speech with Turbo (350M, English, paralinguistic tags like [laugh]), Nano (110M, English, runs on CPU), Multilingual V3 (500M, 23+ languages) or the original English model. Every generated file carries Resemble's Perth watermark.
Who it's for
It suits developers and researchers who want to study or self-run voice cloning models and can handle PyTorch and a GPU. It does not suit anyone needing updates, security fixes or a commercial service: the README says there is no support or paid tier, and Resemble AI does not sell TTS.
Key features
Zero-shot voice cloning
Clone a voice from a roughly 10 second reference clip.
Turbo model with tags
350M English model with native tags such as [cough], [laugh] and [chuckle].
Nano for CPU
110M model the README says runs 3x faster than realtime on 8 CPU cores.
Multilingual V3
500M model for 23+ languages including Arabic, Hindi, Japanese and Swahili.
Single Language Pack
Six dedicated finetunes for priority languages and regional variants.
Built-in Perth watermark
All generated audio includes an imperceptible neural watermark you can extract.
Up and running in 3 steps
Our honest verdict
What's great
- Voice cloning from a short reference clip needs no per-voice training.
- Nano runs on CPU, so you can try it without a GPU.
- Paralinguistic tags in Turbo let you add laughs or coughs inline in the text.
- Output carries a watermark with an extraction script, which helps you label synthetic audio.
Where it falls short
- The README says it is retired and unmaintained: no support, no updates, no security fixes, and no paid tier.
- Resemble AI does not sell voice products and has moved to deepfake detection, so there is no commercial route to this model.
- Turbo and the multilingual models need PyTorch and, for good speed, a CUDA GPU; the README says it was developed on Python 3.11 on Debian 11.
- Because dependencies are pinned in pyproject.toml, it may conflict with newer Python or library versions over time.
Pricing & license
The code is MIT-licensed and free. The README says there is no paid tier and no support; Resemble AI does not sell text-to-speech, so your only cost is the machine you run it on.
Open remote AI-training roles
768 open remote AI-training roles, updated November 14, 2026; most pay $20–$175 an hour. Zekai lists them free; the application goes to the employer’s platform.
Alternatives
These reviewed AI tools do the same job instantly — from our scored directory.
Use it, or have us set it up
FAQ
Is Chatterbox still maintained?
No. The README states that as of August 2026 Chatterbox is retired research and no longer maintained, kept public as an archive under the MIT license.
Does it need a GPU?
The examples use cuda, but the README says Nano targets CPU inference and the original model also lists cpu and mps devices.
Which languages does Multilingual support?
The README lists 23 languages, from Arabic and Danish to Swahili, Turkish and Chinese.
