Chatterbox

Chatterbox is a family of open-source text-to-speech models from Resemble AI. Text-to-speech (TTS) means software that turns written words into spoken audio. You can clone a voice from a short audio sample and tune the emotion with simple settings. It comes in three sizes: Turbo for fast English voice agents, Nano for CPU and on-device use, and Multilingual V3 for 23-plus languages. Install it with one pip command. Check the official GitHub repo for setup steps and the latest release notes.

Key Features

1. Zero-Shot Voice Cloning

Clone any voice from a short reference clip. The docs show a ten-second sample is enough to copy tone, accent, and speaking style.

2. Emotion and Style Control

Tune feeling and pacing with two settings: exaggeration and CFG weight. Higher exaggeration makes the speech sound more dramatic and lively.

3. Three Model Sizes

Turbo (350M) serves fast voice agents, Nano (110M) runs on CPU, and Multilingual V3 (500M) handles 23-plus languages worldwide.

4. Multilingual Speech

The Multilingual V3 model speaks 23 languages, including French, Hindi, Japanese, and Chinese. Cross-language cloning keeps the voice stable.

5. Paralinguistic Tags

Turbo models read tags like [laugh], [cough], and [chuckle] inside your text. They add natural human touches to the finished audio.

6. Built-in Watermarking

Every generated file carries an invisible PerTh watermark. It survives MP3 compression and audio editing, with near-100% detection accuracy.

Frequently Asked Questions (FAQ)

Which Chatterbox model should I use?

Pick Turbo (350M) for low-latency English voice agents, Nano (110M) for CPU or on-device use, and Multilingual V3 (500M) for 23 languages.

How do I install Chatterbox?

Run pip install chatterbox-tts in Python 3.11. The team tested it on Debian 11. GPU helps the larger models; Nano is built for CPU.

How does voice cloning work?

Give the model a short reference clip of the target voice. It copies the tone and style onto your new text. No extra training is needed.

What are paralinguistic tags?

Tags like [laugh], [cough], and [chuckle] that you write inside the text. Turbo models turn them into natural human sounds.

Can it clone a voice in another language?

Yes. Multilingual V3 supports cross-language cloning across 23 languages with steady speaker similarity and fewer speech errors.

Conclusion

Chatterbox gives developers state-of-the-art voice AI under an MIT license. Three model sizes cover fast agents, on-device use, and 23 languages. Voice cloning, emotion controls, and built-in watermarking come ready to use. It suits developers and creators who want voice output without cloud fees. Install it with pip and check the repo docs for the newest model versions.