Kokoro TTS
Python inference library for Kokoro-82M, a small open-weight text-to-speech model with multi-language voices.
Generate speech locally with an 82M-parameter model and Apache-licensed weights.
Kokoro is a Python inference library for the Kokoro-82M text-to-speech model, installable with pip. It suits developers and technically minded creators who want a lightweight voice model they can run on their own hardware and ship in products.
What it is
The kokoro package loads the 82-million-parameter Kokoro-82M model and turns text into 24 kHz audio through a KPipeline. You choose a language code and a voice, such as af_heart, and it yields audio chunks you can save with soundfile. It relies on the misaki G2P library for pronunciation and on espeak-ng for fallback and some languages.
Who it's for
It suits developers adding voiceover or narration to apps, and creators comfortable running Python or a Colab notebook. It is not a finished voice-over studio: there is no graphical editor in this repo, and Japanese and Mandarin need extra misaki installs.
Key features
82M-parameter model
A small TTS model the README describes as faster and cheaper than larger ones.
pip install package
Install with pip install kokoro and run a few lines of Python.
Multiple languages
Codes cover American and British English, Spanish, French, Hindi, Italian, Japanese, Brazilian Portuguese and Mandarin.
Inline pronunciation
Mark words with phoneme syntax like [Kokoro](/kˈOkəɹO/) to fix pronunciation.
Streaming-style output
The pipeline yields graphemes, phonemes and audio per text chunk.
Colab-ready examples
Basic and advanced notebook cells are given in the README.
Up and running in 3 steps
Our honest verdict
What's great
- The model is small at 82 million parameters, so it is far lighter to run than large TTS systems.
- The README says the weights are Apache-licensed, so they can be deployed in production or personal projects.
- Nine language codes are listed, covering several widely spoken languages beyond English.
- You can correct pronunciation inline with phoneme markup instead of retraining anything.
Where it falls short
- It is a Python library, not an app; you write code and manage audio files yourself.
- It needs espeak-ng installed for English fallback and some languages, and Japanese and Mandarin require extra misaki packages.
- The code repo shows no tagged releases or GitHub releases, so you depend on the PyPI package version and the model card for stability.
- The README makes quality claims but offers no side-by-side comparison here; the samples are linked on Hugging Face.
Pricing & license
Free under Apache-2.0, and the README shows no paid plan. Your only cost is the computer or server you run it on.
Open remote AI-training roles
767 open remote AI-training roles, updated November 14, 2026; most pay $20–$175 an hour. Zekai lists them free; the application goes to the employer’s platform.
Alternatives
These reviewed AI tools do the same job instantly — from our scored directory.
Use it, or have us set it up
FAQ
What languages does Kokoro support?
The README lists American and British English, Spanish, French, Hindi, Italian, Japanese, Brazilian Portuguese and Mandarin Chinese.
Can I use it commercially?
The README says the model has Apache-licensed weights and can be deployed from production environments to personal projects; the code repo is Apache-2.0 too.
Do I need a GPU?
The README does not state a hardware requirement; it only describes the model as lightweight at 82 million parameters.
