Kokoro TTS: Open-Weight Text-to-Speech Model
github.com · checked 5 Oct 2026
Open-weight speech synthesis for local, browser, and API workflows at no cost.
How it plugs in
Signal on Text-to-Speech Software7.9#10 of 223
Kokoro TTS uses an 82-million-parameter open-weight model for speech generation. It renders text in eight languages and writes the result as WAV audio. You can run it through an API, browser library, or self-hosted setup on Linux, macOS, Windows, and the web. Its Apache-licensed weights allow commercial and personal projects.

Review
In a line
Kokoro TTS converts written words into spoken audio. Its 82 million parameters are distributed as model weights under Apache licensing.
What it does
The model creates speech from text in eight languages. A Python pipeline lets you select a voice and set the speaking rate.
Generated audio is produced at 24 kHz and saved in WAV format. Kokoro does not provide voice cloning.
How it connects
Kokoro offers API access and can run on Linux, macOS, Windows, a web browser, or your own server. Its JavaScript library runs the model in a browser with Transformers.js and documents CPU use in Node. It can also stream speech while more text arrives.
The project shows Google Colab examples, links to a Hugging Face demo, and lists inference providers and notebooks. The documented Python installation relies on kokoro, soundfile, and eSpeak NG, while Misaki converts graphemes to phonemes. See text-to-speech software for other options.
Good fit
Kokoro suits people who want free text-to-speech for personal or commercial projects and prefer to run the model themselves. Its Apache-licensed weights support production and personal deployment.
It can fit browser projects, API-based applications, and local setups. The project includes guidance for Windows eSpeak NG installation and GPU acceleration on Apple Silicon Macs.
Watch for
Eight languages are supported, and WAV is the only export format listed. Check that these choices match your application before adopting it.
Setup requires several packages, including eSpeak NG, and the model does not clone voices. Running it locally may also require handling the installation and hardware yourself.
Pricing
Kokoro is free. Its listed “Kokoro open-weight model” plan covers 82 million parameters and Apache-licensed weights, and no paid tier is shown.
Verdict
Choose Kokoro if you need no-cost speech generation, API access, or a model you can host in your own environment. It is also a practical choice when commercial use and browser or local operation matter.
Look elsewhere if you need cloned voices, formats beyond WAV, or more than eight languages. Buyers who want a managed service rather than installing packages should compare hosted tools such as Resemble AI or LOVO AI.
Our take
Kokoro TTS fits teams and individuals seeking free speech synthesis that can run locally, in a browser, or behind an API. Apache-licensed weights allow commercial projects, and the model supports voice selection, speed control, and streaming through its documented tools. Choose it when control and zero license cost matter more than a managed service or a wider voice feature set.
- Free model with Apache-licensed weights
- Runs locally, in browsers, through an API, or on your own server
- Supports eight languages with voice and speed controls
- Can stream speech as text arrives
- Commercial use is allowed
- No voice cloning
- WAV is the only listed export format
- Setup requires kokoro, soundfile, eSpeak NG, and Misaki
- Only eight languages are listed
- Pick it ifSelf-hosted speech for personal or commercial projects; Browser, API, and local text-to-speech apps; Users who want free open-weight model files
- Skip it ifYou need voice cloning or audio formats other than WAV; You want a managed service without installing dependencies; Your project requires more than eight languages
Pricing
The model is free, with no paid plan listed. The Apache-licensed weights are available for personal or commercial projects.
- Kokoro open-weight model82 million parameters · Apache-licensed weightsFree
As huggingface.co lists them, checked 5 Oct 2026.
Questions
Is Kokoro TTS free?
Yes. The listed Kokoro open-weight model has no charge and includes Apache-licensed weights.
Can I use Kokoro commercially?
Yes. The maker permits production and personal use of the Apache-licensed weights.
Where can Kokoro run?
It supports API, Linux, macOS, Windows, self-hosted setups, and the web. The kokoro.js library also runs the model locally in a browser.
How many languages does Kokoro support?
Kokoro supports eight languages.
What audio files does it create?
The documented Python example creates 24 kHz WAV files, and WAV is the listed output type.
Does Kokoro clone voices?
No. Voice cloning is not supported.
Can Kokoro stream speech?
Yes. Its JavaScript library describes streaming generated audio as additional text arrives.
What does installation require?
The documented Python setup uses the kokoro package, soundfile, and eSpeak NG. Misaki processes written text into speech sounds.