Kinyarwanda TTS

Most voice AI systems under-serve local languages. They mispronounce our names, flatten our tones, and treat our languages as an afterthought — if they support them at all. Our Kinyarwanda text-to-speech project exists to change that for our language.

Why voice matters

Voice is the most natural interface there is — and the most exclusionary when it doesn't speak your language. For visually impaired users, for new readers, for anyone who thinks in Kinyarwanda, natural voice output is not a convenience. It is access.

How we're building it

There was no off-the-shelf foundation for Kinyarwanda speech synthesis, so every layer had to be built deliberately:

• Dataset curation — carefully collected, speaker-reviewed recordings rather than scraped scraps
• Phonetic analysis — covering the sound system of Kinyarwanda properly, including tone
• Evaluation discipline — naturalness judged by native speakers, not just by metrics

Where it feeds

This research is one pillar of ATAS's language program — alongside speech recognition and natural language understanding that handles how Rwandans actually mix Kinyarwanda, English, and French. The data discipline and evaluation practices developed here flow directly into IMIZI and AcademiaPlus.

Language preservation in the digital age isn't about archiving. It's about building systems where our language lives, works, and evolves.