Kinyarwanda Voice AI

Building speech recognition and synthesis that understands Kinyarwanda the way it's actually spoken.

Elyssa Irankunda

Started September 1, 2025

This research is active and ongoing. We'll publish results and technical detail here as they're validated, not ahead of them.

Kinyarwanda is spoken by over twelve million people, but almost no commercial speech AI supports it well — and the systems that do rarely account for how differently it is spoken across regions, generations, and phone quality. This is one of the foundational research areas ATAS was built around.

What we're working on

Speech recognition that holds up across regional accents and dialects, not just a single "standard" voice

Natural-sounding speech synthesis for Kinyarwanda

Voice interaction that works over basic phones and low-bandwidth connections, not just smartphone apps

Abstract illustration of varied wave patterns converging into a single recognized signal. — A single standard voice model misses how differently Kinyarwanda is actually spoken from place to place.

Interested in collaborating?

Cite this work

ATAS Research Team (2026). Kinyarwanda Voice AI. ATAS Research Center. https://atas.rw/research/kinyarwanda-voice-ai

Project timeline

September 1, 2025
Research area established
Began working on speech recognition and synthesis for Kinyarwanda, starting with dialect and accent coverage.

December 1, 2026Planned
Voice interfaces over USSD
Extending voice interaction to basic phones through USSD, so the work doesn't depend on smartphones or steady internet.