Giving Sukuma a Voice

We believe all languages matter, and that intelligence should live in the languages people actually speak and write—not only the ones that already crowd the internet. After opening Swahili → Sukuma translation and Sukuma speech → text, the missing piece was clear: machines that can speak Sukuma back, not only read or hear it.

Sukuma is spoken by millions, and a language that can be translated and heard should also be spoken back. So today we open the next chapter: two Sukuma text → speech models, shared as a research preview and kept small on purpose so they can run on phones, everyday machines, and places where big cloud GPUs are scarce and the network is thin. Paste Sukuma below and you should hear it come back.

Why two versions?

Not every device has the same room to spare, so we share two sizes. Reach for Quality when you can afford a stronger voice, or Lite when memory and speed have to win. Both are on Hugging Face (quality · lite).

What to expect

This remains a research preview, and the input is one language: Sukuma in macron orthography (ā ē ī ō ū), not English and not raw Swahili. The tone leans literary, so casual chat or domain-heavy wording may sound odd or off-register. Treat the box above as a living demo, and the open weights as a starting point for anyone who wants Sukuma in the loop rather than a finished product for every domain. If you start from Swahili, translate with Sukuma MT first, then bring the Sukuma here—the same path we sketched when we released translation.

Applications

A language that can be translated and heard should also be spoken back. These voices are for literary Sukuma on phones and everyday machines, not every domain at once.

• Read Sukuma aloud

  Paste literary Sukuma and hear it on a device that does not need a cloud GPU, useful for study, proofing, and sharing text with listeners.

• Learning and accessibility

  Support readers who want to hear Kisukuma, and learners who need a spoken check after they write or translate a line.

• A small local loop

  Translate from Swahili, listen with Sukuma speech recognition, then speak the text back—the same path we opened with the earlier previews.

For builders

The try box already does the hard part for you. If you load the weights yourself, prefer the Hugging Face checkpoints linked above and keep requests modest—about 2,000 characters is the ceiling for the live demo. This release sits beside the earlier ones: Swahili → Sukuma (August 20) and Sukuma speech → text (August 22). Together you can sketch a small loop: translate, listen, speak. Methods, scores, and limits are in the technical report below, and in a forthcoming paper.

Collaboration

If you are extending this work, evaluating on new data, or building with Sukuma in your community, write to us at hi@nileagi.com. The weights are released under CC BY-NC-SA 4.0; commercial use needs a written agreement with NileAGI.