Chatterbox TTS

Chatterbox TTS

Resemble AI's open text-to-speech models, with zero-shot voice cloning and emotion exaggeration control. The latest multilingual version covers 23 languages at 0.5B parameters.

Made by ResembleAI on Hugging Face. These notes are Mindket's, written from its model page and model card; the model's own files and license are on Hugging Face.

About

What it is: Chatterbox is a family of open text-to-speech models from Resemble AI, released under the MIT license. It makes speech from text and can copy a voice from a short sample.

The newest release is Chatterbox Multilingual V3, a 0.5B model for 23 languages. The card says it improves speaker similarity, cuts hallucinations and sounds more natural. The family also has a Single Language Pack of 6 dedicated models and the original English Chatterbox.

Who it suits: Developers and creators who want expressive speech for videos, games, apps or agents, and teams that need one voice to speak many languages.

Industry | Who uses it | How they use it | Example
Media | Video creators | Add narration and character voices | Voice a short film in a chosen style
Games | Game developers | Give characters spoken lines | Generate dialogue with extra emotion
Localization | Product teams | Speak one voice in many languages | Make the same message in several languages

Skip it if you cannot get permission to use the voice you want to copy. Only use faces, voices and recordings you have the right to use.

Where to use it:

• A Python setup that follows the model card
• The demo pages the makers list for each model, if you want to listen first
• Windows, macOS or Linux with Python, to download the files

Use case | Who | What they get back
Narration | Creators | Speech from text in a voice you choose
Cross-language voice | Localization teams | A voice that keeps its identity and accent across languages
Expressive reads | Game and media makers | Speech with more or less emotion, through the exaggeration control

How to set it up:

• Python and pip
• Enough disk space for the model files, at about 500M parameters per model
• A short recording of a voice you have the right to use, if you want cloning

1. Install the Hugging Face tools.
2. Download the model files.
3. Follow the model card for the code that loads the model and generates speech.

pip install -U huggingface_hub
hf download ResembleAI/chatterbox

How to use it: Give the model your text, and optionally a short voice sample to copy. The original model also offers CFG and exaggeration settings to make the delivery calmer or more dramatic.

Model | Size | Languages | Best for
Chatterbox Multilingual V3 | 500M | 23+ | Global apps and cross-language voice cloning
Single Language Pack | 500M each | 6 dedicated finetunes | Priority languages and dialect-sensitive work
Chatterbox | 500M | English | General zero-shot speech with creative controls

Title: Languages per Chatterbox model
Chatterbox original: 1 languages
Single Language Pack: 6 languages
Multilingual V3: 23 languages

Example: You give the model a line of text and a short recording of your own voice. It speaks the line in a voice like yours, and you adjust the exaggeration setting to make it more lively.

For newcomers and experts: If you are new to speech models, start with the English model and plain text before you try cloning. Then move to the multilingual version.

If you know the field, V3 keeps the 0.5B size of earlier versions and focuses on speaker similarity and stability. The card says it reduces repeated and off-prompt speech.

Download the model files
Load the model in Python
Give it text and an optional voice sample
Receive speech

Good to know:

• The makers say it was compared with closed systems such as ElevenLabs and often preferred, and these claims are theirs
• The card calls it the first open source TTS model with emotion exaggeration control
• The makers also mention a paid hosted service with latency under 200 ms
• Hugging Face lists 40 quantized and 93 fine-tuned versions made by other people
• License: MIT
• Cost to run: none named for the open models
• Data it reaches: the files you download, the text you give it and any voice samples you provide
• Not for: copying a voice without the right to use it

On Hugging Face, 13 November 2026

• Task: text to speech
• Library: chatterbox
• Downloads in the last 30 days: 2,440,902
• Likes: 1,911
• License: mit
• Languages: ar, da, de, el, en, es, fi, fr
• Quantized versions: 40
• Fine-tunes: 94
• Tags: chatterbox, text-to-speech, speech, speech-generation, voice-cloning, multilingual-tts

Links

• Model: https://huggingface.co/ResembleAI/chatterbox
• Files: https://huggingface.co/ResembleAI/chatterbox/tree/main

Get the model

Download its files with Hugging Face's command line:

pip install -U huggingface_hub
hf download ResembleAI/chatterbox

Then follow the usage in its model card. Nothing in this kit runs by itself.