Skip to content

Chatterbox AI

An open production-ready TTS model with emotion control

Visit official site

What Chatterbox AI is

The model is described as the first open-source production-grade text-to-speech model from Resemble AI, using zero-shot synthesis on a 0.5B Llama backbone. It offers emotion exaggeration control, aligned inference and low latency.

What you can do with it

  • Generate natural speech from different inputs with zero-shot TTS
  • Adjust emotion exaggeration and intensity in the voice
  • Rely on aligned inference for fluent speech
  • Use the built-in watermark on generated audio

Application scenarios

  • Produce voice dialogue for game characters
  • Add emotionally rich narration to a video
  • Create an AI assistant with a personalised voice

Getting started

  • Install the library with pip install chatterbox-tts
  • Import torchaudio and the ChatterboxTTS module in Python
  • Load the model with from_pretrained and set the device to cuda
  • Call generate on the text to produce audio data

Pros & cons

✓ What we like

  • Training uses 500,000 hours of cleaned data
  • A watermark is built into generated audio

! What to watch out for

  • The documented load step specifies a CUDA device
  • Using the model requires installing the chatterbox-tts package

FAQ

What kind of model is it?

The material describes an open-source production-grade TTS model from Resemble AI.

How is speech generated?

The documented steps load the model with from_pretrained and call generate on the input text.

Last reviewed: 2026-11-15

More AI audio tools tools

View all →

How we review