Chatterbox

// Overview

Chatterbox is a family of open-source text-to-speech (TTS) models from Resemble AI. You give it text and a short reference clip of a voice, and it generates speech that matches that voice. It runs on a single GPU, so you can use it locally without a cloud service.

The family covers different needs. Chatterbox-Turbo is a smaller 350M-parameter model aimed at low-latency English voice agents, and it supports paralinguistic tags like [laugh] and [cough] for more realistic speech. Chatterbox Multilingual V3 is a 500M model that handles 23+ languages with steadier speaker similarity across them, and there are single-language finetunes for priority languages.

It fits the speech and audio space for developers who want zero-shot voice cloning they can run themselves, for example in voice agents, narration, or localization workflows. A hosted Resemble AI service is available if you later need to scale beyond local hardware.

// What it does

• Zero-shot voice cloning from a short (around 10s) reference clip
• Multilingual generation across 23+ languages with Chatterbox Multilingual V3
• Chatterbox-Turbo: a 350M model for low-latency English voice agents
• Paralinguistic tags such as [cough], [laugh], and [chuckle] (native in Turbo)
• Runs on a single GPU, with device options for cuda, cpu, or mps
• Installable from PyPI (chatterbox-tts) or from source

// Getting started

Install the package from PyPI, then load a model and generate speech from text plus a reference voice clip.

pip install chatterbox-tts

git clone https://github.com/resemble-ai/chatterbox.git
cd chatterbox
pip install -e .

import torchaudio as ta
from chatterbox.tts import ChatterboxTTS

device = "cuda"  # or "cpu" / "mps"
model = ChatterboxTTS.from_pretrained(device=device)

wav = model.generate("Hello there!", audio_prompt_path="your_10s_ref_clip.wav")
ta.save("test.wav", wav, model.sr)

import torchaudio as ta
from chatterbox.tts_turbo import ChatterboxTurboTTS

model = ChatterboxTurboTTS.from_pretrained(device="cuda")
text = "Have you got one minute to chat [chuckle]?"
wav = model.generate(text, audio_prompt_path="your_10s_ref_clip.wav")
ta.save("test-turbo.wav", wav, model.sr)

Commands and code are distilled from the project's own documentation — always check the official repo for the latest.

// When to use it

• Build a low-latency English voice agent that reads back responses in a chosen voice
• Clone a voice from a short clip to narrate articles, videos, or audiobooks
• Localize spoken content across many languages while keeping a consistent speaker identity
• Add expressive cues like [laugh] or [cough] to make generated speech sound more natural

// How Chatterbox compares

Chatterbox alongside other open-source audio, music & voice tools AI/TLDR tracks, ranked by GitHub stars.