Hearing Kiswahili

Millions of people speak Swahili every day. Too often their voices only become usable to software after someone routes them through another language first. We think that path is backwards: speech technology should meet people in the language they already use.

Today we are sharing a research preview of Swahili speech → Swahili text. The public checkpoint is nileagi/nileagi-sw-stt. It does one job. Record or upload clear Swahili and read the transcript back below.

What to expect

This is a research preview of Swahili speech recognition. Record or upload clear, single-speaker Kiswahili and you should get a readable transcript back. Telephone audio, heavy noise, overlapping speakers, and long stretches of silence are outside what we treat as the intended setting.

Short, clean clips work best in the live demo. Prefer 16 kHz mono audio, and keep a clip to about half a minute when you can. Review anything that has to stand as a human record.

Applications

This is a research preview. The useful setting is clear, single-speaker Kiswahili that should become text without being routed through English first.

• Radio, interviews, and archives

  Turn spoken Swahili programmes and conversations into searchable notes, with a human still reviewing anything that has to stand as a record.

• Classrooms and offices

  Capture lessons, meetings, and dictation in the language people already use, then edit the transcript instead of typing from scratch.

• Access without typing

  Let people speak Kiswahili to a tool on a phone or everyday machine when a keyboard is slow, unavailable, or the wrong interface.

For builders

Load the model from Hugging Face with Transformers, keep audio at 16 kHz mono, and treat clear research transcription as the intended use. Sample audio ships in the Hub repository and in the try box above. The weights follow the MIT License that accompanies the public checkpoint.

Collaboration

If you are evaluating on new Swahili speech, adapting this checkpoint for your setting, or exploring what comes after transcription, write to hi@nileagi.com. Use this system for clear Swahili research transcription, review anything that has to stand as a human record, and keep legal, medical, or emergency use out of scope for now.