Chatterbox

Chatterbox is a family of open-source text-to-speech (TTS) models from the voice AI company Resemble AI on GitHub. You install and call it from Python, and it generates speech on your own computer. Give it text and it reads it aloud; add a reference clip of about 10 seconds and it reads in that voice, which is voice cloning. The README lists several models. Chatterbox-Turbo (350M parameters, English) lets you write tags such as [laugh] and [cough] into the text to add laughs and coughs. Chatterbox-Nano (110M parameters) shares Turbo's architecture, and the README says it runs 3x faster than real time on 8 CPU cores. Multilingual V3 (500M parameters) covers 23 languages including Chinese, Japanese and Arabic, with 6 single-language versions such as Mandarin Chinese, Latin American Spanish and Hindi. The repo also includes a voice conversion example that turns one recording into another voice. What sets it apart is that, according to the README, every generated audio file carries Resemble AI's imperceptible PerTh watermark, and a script is included to detect it, so the audio can later be identified as AI-generated.

Who it's for:developers and creators who know some Python and want narration for videos, audio content or voice assistants made on their own computer, using their own voice or a reference recording they have permission to use. Readers who don't write code will find it hard going.

Before you start: get the person's consent before you use someone else's voice. Resemble AI's consent pledge on its website says: "No permission? No voice cloning." That is a company commitment; the open-source code itself doesn't check whether you have permission, and the MIT license doesn't give you the right to clone anyone's voice. Install with pip install chatterbox-tts; the authors developed and tested it on Python 3.11 and Debian 11. The first run downloads model weights from Hugging Face, and it runs on NVIDIA GPUs (CUDA), Apple silicon (MPS) or the CPU, with speed depending on your hardware. The code is MIT licensed; check each model's card on Hugging Face for the license of its weights. In local use, your reference recording is processed on your own computer. If you use the Hugging Face online demos linked in the README, your recording is uploaded to the demo page. The README also promotes Resemble AI's paid voice service, but a notice on Resemble AI's website now says the company has stopped offering voice generation as a commercial product, and that the voice models it built remain open and free to use for research and development. The README's closing disclaimer reads: "Don't use this model to do bad things."

Based on the project's public documentation and Resemble AI's website. Not tested by AI Inspiration World.