Is Kokoro TTS free for commercial use?

Yes. The Kokoro-82M weights are licensed under Apache 2.0, and the authors welcome commercial deployments. Supertonic 3 is royalty-free too, under OpenRAIL-M with a list of prohibited uses; voices bought on SuperClonica are for your own projects and may not be resold.

Can Kokoro clone a voice?

Not with the official release: it comes with 54 preset voices, and no encoder was published. To get a voice from your own recording, train a Supertonic 3 voice on SuperClonica or commission one from an author. Use only recordings you have rights to, with the speaker's consent.

Does Kokoro speak Russian or German?

No. Kokoro v1.0 covers English, Japanese, Mandarin Chinese, Spanish, French, Hindi, Italian and Brazilian Portuguese. Supertonic 3 voices speak Russian, German, Korean and 28 other languages, but not Chinese.

Is SuperClonica faster than Kokoro?

We have not measured Kokoro, so we cannot say. Our estimate for SuperClonica is 6.4–9.7× faster than real time on a desktop 8-core CPU. Both can run without a GPU, so the fairest test is your own text on your own machine.

MODEL COMPARISON

Kokoro TTS vs SuperClonica: two small models for local speech

Updated October 23, 2026

Kokoro-82M is an open-weight text-to-speech model with 82 million parameters, released under Apache 2.0. Version 1.0 came out in January 2025 with 54 preset voices in eight languages. Supertonic 3, the model behind SuperClonica, is in the same weight class: about 99 million parameters, made to run on an ordinary CPU.

So the choice is less about size and more about languages, voices and licenses. Kokoro is not in our comparison table and we have not benchmarked it, so this page gives no speed or quality scores for it, only facts from its model card and repository.

Hear voices in 31 languages Install SuperClonica

What Kokoro-82M is

Kokoro is built on StyleTTS 2 with an ISTFTNet decoder and was trained on a few hundred hours of public-domain, permissively licensed and synthetic audio. The weights are under Apache 2.0, and the authors explicitly welcome deployments in commercial products.

You install it with pip install kokoro and add espeak-ng, which handles words missing from the English lexicon and some non-English languages. The output is 24 kHz audio. A separate JavaScript library, kokoro-js, runs the model entirely in the browser through Transformers.js.

Kokoro languages compared with Supertonic 3

Version 1.0 has 54 voices: 20 American English, 8 British English, 8 Mandarin Chinese, 5 Japanese, 4 Hindi, 3 Spanish, 3 Brazilian Portuguese, 2 Italian and 1 French. Its own voice list warns that support for non-English languages may be thin because of weaker phonemization and less training data. It grades the Mandarin voices D and the Japanese ones between C+ and C-.

Supertonic 3 speaks 31 languages, among them Russian, German, Korean, Polish, Ukrainian and Turkish, which Kokoro does not cover at all. It covers every Kokoro language except Chinese: if you need Mandarin, Kokoro has voices and Supertonic 3 does not.

Preset voices or a voice of your own

Kokoro ships with its preset voices. The model card notes that no encoder was released, and the official package has no way to turn your recording into a new voice.

SuperClonica is built around custom voices. The catalog holds Supertonic 3 voices from independent authors, each with demos and a price in US dollars, and some are free. If nothing fits, commission a voice from a reference recording, with the budget reserved on your balance until you accept the result, or train one yourself for free from a clean 5- to 10-second recording on your own NVIDIA GPU. Either way you get one JSON file that runs offline.

Speed on an ordinary CPU

We have not measured Kokoro, so there are no speed figures for it here. At 82 million parameters it is slightly smaller than Supertonic 3; test it on your own hardware with your own texts.

For SuperClonica we estimate 6.4–9.7× faster than real time on a desktop 8-core CPU, so a 10-second sentence takes about 1.0–1.6 seconds. The engine is on average 61% faster than the standard Supertonic 3 client on CPU, voices long texts in short parts so words are skipped or repeated less often, and reads the same voice files without conversion.

When Kokoro is the better choice

Pick SuperClonica when you need a language Kokoro lacks, a voice beyond the presets, or a voice trained from your recording. Kokoro makes more sense in these cases:

You want a free, small English TTS model and the preset voices are enough.
You need Mandarin Chinese, which Supertonic 3 does not have yet.
You want the most permissive terms: Apache 2.0 has no list of prohibited uses, unlike OpenRAIL-M.
Your app runs speech in the browser and the kokoro-js package already fits your stack.