What Is Chatterbox? Turkish Text-to-Speech, Voice Cloning, Turbo vs Nano, License and Watermark

You install Chatterbox on your own computer, give it text and get an audio file back; it also clones a voice from a short recording. It's Resemble AI's MIT-licensed, open-source speech synthesis (text-to-speech) and voice cloning model, and it requires some Python knowledge. Turkish is only available in the 23-language (Multilingual) model; Turbo and Nano are English only. Resemble's Perth watermark is added to every output. There's no official benchmark for Turkish quality; there's an open issue on GitHub reporting that Turkish letters are skipped.

What it does: Turns text into speech, clones voices
Free: Yes, MIT; your own hardware
Turkish: Only in the multilingual model; quality unclear
No-code: No; requires Python
Watermark: Perth watermark on every output

What Is Chatterbox For? Can I Use It Without Coding?

You give Chatterbox some text and it produces spoken audio. If you give it a voice recording a few seconds long, it can also produce speech that sounds like that voice (voice cloning). It's not an app but a Python library: the model files download from Hugging Face and run on your own computer.

Voice Turkish textWhat you get: WAV audioNeeds a separate setup Clone a voice from a recordingWhat you get: WAV audioNeeds a separate setup

Can I use it without coding?

In practice, no: installation and use are done with Python commands. A Hugging Face demo link in the README lets you try it in the browser, but that's for Turbo (English).

Who it's for: developers who know Python and want to experiment with voiceovers or voice agents, or generate speech on their own hardware.

How Do You Install Chatterbox? What Hardware Do You Need?

Set up an environment with Python 3.10 or later (the README example uses 3.11).
Install the package: pip install chatterbox-tts
Load the model with ChatterboxMultilingualTTS.from_pretrained(device="cuda") and generate Turkish with generate(text, language_id="tr").
You can choose cuda, cpu or mps (Mac) as the device.

There's no official minimum VRAM or RAM figure. A community issue mentions about 6-7 GB of VRAM; that's the only source. The reference clip's language should match the selected language, or a foreign accent may come through (the README suggests cfg_weight=0). There's no official statement of support for Windows and Mac.

Which Model? Turkish, Turbo and Nano Compared

Model	Size	Language	What it's for, according to the README
Multilingual (V3)	500M	23 languages, including Turkish	General-purpose multilingual model; recommended for multilingual cloning
Chatterbox-Turbo	350M	English only	Low-latency voice agents; tags like [laugh] and [cough]; less VRAM
Chatterbox-Nano	110M	English only	CPU and on-device; 3x faster than real time on 8 cores
Single-language packs	500M each	Chinese, Spanish (2), Portuguese (2), Hindi	Language-specific fine-tuning; no pack for Turkish

Turkish voiceoverPartly	Only in the multilingual model. An open GitHub issue (#261) reports that the Turkish letters ç, ş, ü, ğ and ö are skipped or mispronounced; whether this was fixed in V3 is not clear from the official documentation.
A separate model for TurkishNo	The single-language packs don't include Turkish; an issue requesting a separate Turkish model (#541) is open.

The speed and quality statements in the README are the maker's own claims. This section was checked on October 14, 2026.

License, Watermark and Commercial Use

License

MIT: commercial use, modification and distribution are allowed; the copyright notice must be kept. There's no separate clause on the rights to the generated audio.

Perth watermark

The README says Resemble AI's inaudible Perth watermark is added to every output, and the code applies it in the generation pipeline. No setting to turn it off is documented. “Nearly 100% detection” is the maker's claim.

Consent

Taking a sample of someone else's voice may involve personal data; consent and compliance with data protection law (in Turkey, KVKK, the Personal Data Protection Law) are your responsibility. There's no official privacy statement about whether text or audio is sent anywhere when running it locally.

Frequently asked questions

What is Chatterbox, and what is it for?

Resemble AI's MIT-licensed, open-source speech synthesis and voice cloning model. It turns text into speech and produces speech that sounds like a short voice recording.

Does Chatterbox work in Turkish?

Turkish is only available in the 23-language multilingual model. Turbo and Nano are English only. There's no official quality benchmark; an open issue reports that Turkish letters are skipped.

What's the difference between Chatterbox Turbo and Nano?

Both are English. Turbo, with 350M parameters, is for low-latency voice agents; Nano, with 110M parameters, is for CPU and on-device use.

Is Chatterbox free, and can it be used commercially?

Yes, it's MIT-licensed; commercial use is allowed. You must secure the rights to the reference voice and the person's consent.

Is there a watermark in Chatterbox output?

Yes, the README says Resemble AI's Perth watermark is added to every output. No setting to turn it off is documented.

What computer do you need for Chatterbox?

There's no official minimum. A community issue mentions about 6-7 GB of VRAM. Nano is designed to run on a CPU.

Tasks and setup kits that use Chatterbox

Creating a voiceover for an ad or video

Related guides

Deepgram EMA Lightning

Sources

Not clear in the official sources: Minimum hardware · the state of Turkish quality in V3 · whether the watermark can be turned off with a setting · pricing and privacy rules for Resemble AI's hosted service · the rights status of the output audio · official support for Windows and Mac. Each one is covered in its section, along with what you can do.