Text to speech and voice cloning

Chatterbox

A family of text-to-speech models from Resemble AI under MIT, with zero-shot voice cloning from a reference clip and a multilingual model covering 23 languages.

OSIAlternative to ElevenLabs

What it does

Chatterbox is a family of text-to-speech models published by Resemble AI. The README describes five models: Chatterbox-Turbo (350 million parameters, English), Chatterbox-Nano (110 million, English), Chatterbox-Multilingual V3 (500 million, 23 languages), a set of six single-language finetunes at 500 million each, and the original 500 million English model.

Voice cloning is built in. You pass a reference clip, and the README's examples use a ten-second file, so no training step is needed. The multilingual model lists Arabic, Danish, German, Greek, English, Spanish, Finnish, French, Hebrew, Hindi, Italian, Japanese, Korean, Malay, Dutch, Norwegian, Polish, Portuguese, Russian, Swedish, Swahili, Turkish and Chinese.

The Turbo and Nano models accept paralinguistic tags written into the text, such as [cough], [laugh] and [chuckle]. Every generated file carries Resemble's Perth watermark, and the README includes a script to detect it.

What it replaces

Chatterbox replaces ElevenLabs for voice generation and voice cloning where the licence has to permit commercial use. The code is MIT and the weights on Hugging Face are MIT, which is the reason it sits at the top of the ElevenLabs list on this site.

Where it falls short: the README does not mention streaming, and it does not state a sample rate. ElevenLabs' own claim is 70+ languages against Chatterbox's 23. Voice quality is not compared here, because nobody has measured both on the same test. The demo page is linked above so you can listen with your own text.

How it runs

pip install chatterbox-tts, or clone the repository and install in editable mode. It is a Python library, so serving it over HTTP means writing a small wrapper. The examples pass a device of cuda, cpu or mps. The README says Nano "also runs on CPU" and describes it "running 3x faster than realtime on 8 CPU cores". Development and testing were on Python 3.11 on Debian 11; Windows and macOS are not named.

Resemble AI sells a hosted service, and the README points to it "if you like the model but need to scale or tune it".

Who it fits

A team that wants voice cloning it can ship in a product, with a licence it does not have to negotiate. A team that needs many languages beyond the 23 listed, or a streaming API without writing one, should read the other entries on the ElevenLabs page.

Limits the README states

The reference clip should match the language tag, or the output "may inherit the accent of the reference clip's language". A higher exaggeration setting "tends to speed up speech". The latest tagged GitHub release is from June 2025; the README's own "latest release" heading refers to the Multilingual V3 model, which has no tag.

Sources

More text to speech and voice cloning

ProjectReplacesOpennessStarsLast releaseSelf-host
Fish Speech

A multilingual text-to-speech system with voice cloning from a 10 to 30 second clip and inline emotion tags, released with its weights under a research licence that requires a separate agreement for commercial use.

Source available
32,719
May 31, 2025
Yes
F5-TTS

A research text-to-speech model with zero-shot voice cloning whose code is MIT but whose pretrained weights are non-commercial, because of the dataset they were trained on.

Open weights
15,239
Jul 23, 2026
Yes
Piper

A fast, local neural text-to-speech engine maintained by the Open Home Foundation, used by Home Assistant and the NVDA screen reader, under GPL-3.0.

OSI
5,598
Sep 4, 2026
Yes
Kokoro FastAPI

A Docker image that serves the 82-million-parameter Kokoro model through an OpenAI-compatible speech endpoint, with streaming, voice mixing and nine languages, all under Apache-2.0.

OSI
5,449
Sep 10, 2026
Yes

Added September 16, 2026. Found an error on this page? Write to reveneau@licheo.com and it gets fixed in the next weekly pass.

Running one of these inside your own environment, with your own data and your own security rules, is the kind of work Reveneau does. Read how a forward deployed engagement works.