Supertonic: offline voices in 31 languages

Short answer: Supertonic is an on-device text-to-speech model that Lectern now ships next to Piper and Kokoro. One download of about 129 MB reads 31 languages with ten voices, and it runs comfortably on mid-range phones. It has been checked by ear for English and Dutch so far; in the other languages it works, but names and numbers can come out wrong.

What Supertonic is

Supertonic is a speech model made by Supertone Inc. and published with open weights under the OpenRAIL-M licence. Lectern runs it through sherpa-onnx, the same on-device runtime it uses for Piper and Kokoro, so nothing is sent anywhere: the text goes in on your phone and the audio comes out on your phone. Each download carries ten voices, five female and five male.

The reason it’s in Lectern is the gap it fills. Piper is light and covers many languages, with one voice per download. Kokoro sounds very natural but weighs about 350 MB, reads eight languages and wants a fast phone. Supertonic sits between them: far more languages than Kokoro, more voices per download than Piper, and light enough for the phones most people actually own.

Where it fits among Lectern’s voices

Three downloads, pick the one that matches your books

• English, about 85 MB. Only English, and the clearest English Supertonic has.
• Five languages, about 85 MB. English, Korean, Spanish, Portuguese and French.
• 31 languages, about 129 MB. English, Arabic, Bulgarian, Croatian, Czech, Danish, Dutch, Estonian, Finnish, French, German, Greek, Hindi, Hungarian, Indonesian, Italian, Japanese, Korean, Latvian, Lithuanian, Polish, Portuguese, Romanian, Russian, Slovak, Slovenian, Spanish, Swedish, Turkish, Ukrainian and Vietnamese.

Supertonic doesn’t read Chinese, Norwegian or Hebrew. For Chinese, Kokoro does.

When two downloads both read a language, Lectern picks the one that reads it best: for English that’s the English download, which sounds clearer than the 31-language one. You can change that under Read by for each language if you prefer the other.

The honest limits

• Not every language has been listened to. English and Dutch have had a proper listening round. In the other languages Lectern marks Supertonic as Not yet reviewed, because names, numbers and abbreviations may be mispronounced. Lectern already rewrites numbers, times and common abbreviations before Supertonic reads them, but a language nobody has checked by ear deserves that label.
• It still needs some processing power. Supertonic generates speech on the phone while you listen. On a slower phone it can fall behind, and then narration stutters or pauses. Lectern measures how fast your phone keeps up and says so in the voice list (“may stutter”, “likely to stall”). It warns, it doesn’t forbid: if you like the voice, you can still use it.
• A licence with use restrictions. OpenRAIL-M lets anyone use the voices, but not to deceive, harass or harm people, or to impersonate someone without consent. Lectern shows that once, before the first download.

How to try it

1. Open a book and open the reader settings.
2. Go to the Audio tab and open Voice. The list shows every engine that can read the book’s language, each with a quality word for that language and its download size.
3. Tap Supertonic, download it, and tap play next to a voice to hear it.
4. Pick a voice. It now reads all your books in that language, at up to 3x speed, with the spoken sentence highlighted as it goes.

If you’d rather not choose at all, leave it on Auto: Lectern picks the best voice your phone can keep up with for each language. How Auto chooses, and when to pick yourself.

Frequently asked