- Home
- Text-to-speech tools
- Kokoro TTS
Kokoro TTS
What it is
Kokoro TTS is a free, open-weight text-to-speech model with 82 million parameters and Apache-licensed weights. It uses a decoder-only design built with StyleTTS 2 and ISTFTNet and supports eight languages. The maker describes it as lightweight and more efficient than larger models while providing comparable quality. A Python pipeline lets users select a voice and speech speed, then creates 24 kHz WAV audio. Kokoro.js runs the model locally in a browser through Transformers.js, with Node CPU use documented, and can stream speech as text is added. Commercial use is listed as permitted for personal or production deployment. Python setup uses the kokoro package, soundfile, and espeak-ng, while Misaki handles grapheme-to-phoneme processing. Documentation includes Windows guidance for espeak-ng and Apple Silicon GPU instructions for macOS M1 through M4. Examples point to Google Colab and a Hugging Face demo, with a Discord community linked.
For and against
For
- Free Apache-licensed model weights
- Supports eight languages
- Browser and Node execution documented
- Speech speed and voice selection
- Commercial use permitted
Against
- Outputs are documented as WAV files
- Python setup requires listed dependencies
- Voice cloning is not supported
The verdict
Kokoro TTS provides an open-weight, locally deployable speech model with browser, Node, and Python paths. Its documented setup and output format should be checked against the needs of a particular project.
Who it suits
Kokoro TTS suits developers needing locally runnable speech generation, browser or Node use, and commercial deployment under Apache-licensed weights. It is also relevant to projects requiring selectable voices, speed control, and WAV output.
Plans and prices
Kokoro open-weight model
82 million parameters · Apache-licensed weights
Read 5 Oct 2026 · huggingface.co
On the record
- API access
- Yesgithub.com · read 20 Sep 2026
- Architecture
- The model uses StyleTTS 2 and ISTFTNet and is decoder-only.huggingface.co · read 5 Oct 2026
- Audio output
- The Python example generates audio at 24 kHz and saves it as WAV files.github.com · read 5 Oct 2026
- Commercial use
- Yesgithub.com · read 20 Sep 2026
- Dependencies
- The documented Python installation uses the kokoro package, soundfile, and espeak-ng; Kokoro uses Misaki for grapheme-to-phoneme processing.github.com · read 5 Oct 2026
- Efficiency
- The maker describes Kokoro as lightweight, faster, and more cost-efficient than larger models while delivering comparable quality.huggingface.co · read 5 Oct 2026
- Export formats
- WAVgithub.com · read 20 Sep 2026
- Integrations
- The maker provides examples using Google Colab and points to a Hugging Face demo; the model page also lists inference providers and notebooks.github.com · read 5 Oct 2026
- JavaScript library
- The kokoro.js library runs the model locally in a browser through Transformers.js and also documents Node CPU use.github.com · read 5 Oct 2026
- Languages
- 8github.com · read 20 Sep 2026
- Languages
- The usage documentation lists American English, British English, Spanish, French, Hindi, Italian, Japanese, Brazilian Portuguese, and Mandarin Chinese.github.com · read 5 Oct 2026
- License
- The model weights are Apache-licensed, and the maker says Kokoro can be deployed in production or personal projects.huggingface.co · read 5 Oct 2026
- Platform guidance
- The README gives Windows installation guidance for espeak-ng and Apple Silicon GPU acceleration instructions for macOS M1 through M4.github.com · read 5 Oct 2026
- Purpose
- Kokoro is an open-weight text-to-speech model with 82 million parameters.huggingface.co · read 5 Oct 2026
- Streaming
- The JavaScript library documents streaming generated speech as text is added over time.github.com · read 5 Oct 2026
- Support
- The repository links to a Discord community for Kokoro users and contributors.github.com · read 5 Oct 2026
- Training data
- The model card says Kokoro was trained on permissive or non-copyrighted audio data and IPA phoneme labels, with a dataset of a few hundred audio hours.huggingface.co · read 5 Oct 2026
- Voice and speed controls
- The Python example selects a voice and sets speech speed through the pipeline.github.com · read 5 Oct 2026
