Kokoro-82M review

What is Kokoro-82M?

Kokoro-82M is an open-weight text-to-speech model with 82 million parameters. Version 1.0 has 54 voices across 8 languages. The weights are free under the Apache 2.0 license, and you run it yourself with the kokoro Python package on Windows, macOS or Linux. There is no account, no subscription and no company server in the loop. The model card lists no content policy or usage limits, so the model reads whatever text you give it. That also means nobody checks your age and nothing stops misuse. It does one job: turn text into speech on your machine.

Key features

Fully offline

Everything runs on your own hardware. Your text and the audio never leave your computer. For adult audio, that removes the training and retention questions that come with cloud tools.

54 voices in 8 languages

The voice set is small next to cloud libraries, but it covers several languages, and 54 options give you room to find a narrator you like.

Free for commercial use

The Apache 2.0 license allows commercial use. If you want to narrate stories for a project, there is no fee or revenue share to plan around.

Small and quick

With 82 million parameters, the model is small next to most modern speech models. Small means fast, and a quick model is easier to live with when you run it at home.

The catch

You need to be comfortable with Python. There is no app with buttons. The model card does not mention emotion control, so delivery is plain, and there is no cloning. There are no live calls. With no safety layer, all the responsibility for what you make sits with you. There are no public reviews to judge long-term use.

Final verdict

If you want adult text read aloud for free and in full privacy, Kokoro is hard to beat on price and privacy. You give up expressive delivery, cloning and any kind of conversation. It is a good base for a local project, and a poor fit if you want something that works out of the box.