Chatterbox-Turbo

Chatterbox-Turbo

Speaks English text aloud with a 350M-parameter model built for low-latency voice agents, and adds laughs and coughs through tags in the text. Suits developers who want quick, expressive speech with less GPU memory.

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-Turbo is an open-source text-to-speech model from Resemble AI, one of three models in the Chatterbox family. It has about 350 million parameters and speaks English. The maker cut the mel decoder from 10 steps to one, so it needs less compute and video memory than earlier models.

Who it suits: Developers who build voice agents, narration or other speech features and want expressive output, under the MIT license.

Industry | Who uses it | How they use it | Example
Voice products | Agent builders | Give a voice agent fast, natural speech | Call back a customer with a spoken reply
Media | Content creators | Narrate scripts with laughs and coughs | Add a chuckle to a line
Software | App developers | Add spoken output to an app | Read a message aloud
Customer service | Support teams | Test a phone-style voice assistant | Prototype a billing call

Skip it if you need languages other than English, since the Multilingual model in the same family covers more than 23.

Where to use it: It runs in Python, and the card's example loads it on a CUDA GPU.

• Python, with the chatterbox-tts package from pip
• A computer with a CUDA GPU, as in the card's example
• Windows, macOS or Linux, though the maker tested on Debian 11 with Python 3.11

Use case | Who | What they get back
Voice agents | Agent builders | Low-latency spoken replies
Narration | Creators | Speech with tags marking laughs, coughs and chuckles
App speech | Developers | Audio files made from text

How to set it up: You need Python and the chatterbox-tts package.

• Python 3.11 is what the maker used for development and testing
• pip, to install the package
• A GPU for the card's example, which loads the model with the cuda device

1. Install the chatterbox-tts package with pip.
2. Or install it from source in editable mode, which lets you change the code.
3. Download the model files with the hf command if you want them on disk.

pip install chatterbox-tts

pip install -U huggingface_hub
hf download ResembleAI/chatterbox-turbo

How to use it: Import the Turbo class, load the model, then pass it text that may hold paralinguistic tags. The card begins its example like this.

import torchaudio as ta
import torch
from chatterbox.tts_turbo import ChatterboxTurboTTS

model = ChatterboxTurboTTS.from_pretrained(device="cuda")

Tool | What it does
ChatterboxTurboTTS | Loads and runs the Turbo model
from_pretrained | Fetches the weights and puts them on the chosen device
[laugh] | Tag that adds a laugh to the speech
[cough] | Tag that adds a cough
[chuckle] | Tag that adds a chuckle

Example: You write a short English line for a phone call and place a chuckle tag in the middle. The model speaks the line as audio, and the chuckle lands where you put the tag.

For newcomers and experts: If you are new to text to speech, install the package, load the model and say one short sentence. Then add a tag to hear how it changes the delivery.

If you know the field, Turbo uses a lean 350M design. The speech-token-to-mel decoder was distilled so generation takes one step instead of ten, while the maker says audio fidelity stays high. Tags are native to the model, which is built mainly for low-latency agents.

Install chatterbox-tts
Load the Turbo model
Write text with optional tags
Generate speech
Save or play the audio

Good to know: The family also has Multilingual and the original Chatterbox, and the card says Multilingual offers zero-shot cloning. Use only voices you have the right to use.

Title: Model size in the Chatterbox family
Turbo: 350 M
Multilingual: 500 M
Chatterbox: 500 M

• License: MIT
• Cost to run: free on your own machine, though the card also promotes a paid text-to-speech service from Resemble
• Data it reaches: the text you give it and the model files on your computer, and nothing else the card names
• Not for: languages other than English, since Turbo is English only

On Hugging Face, 13 November 2026

• Task: text to speech
• Downloads in the last 30 days: 0
• Likes: 709
• License: mit
• Languages: en
• Quantized versions: 22
• Fine-tunes: 13
• Tags: text-to-speech, speech, speech-generation, voice-cloning

Links

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

Get the model

Download its files with Hugging Face's command line:

pip install -U huggingface_hub
hf download ResembleAI/chatterbox-turbo

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