In research
Aïr
A voice agent that holds a conversation in Hausa. It listens while it speaks, so it can wait, stop or acknowledge.
People answer before a sentence ends, say “to” while the other person is still speaking, and stop when interrupted. Aïr is our research into speech that does the same, in Hausa.
Listen A greeting in Aïr’s voice, with each word lit as it is spoken.
How it works Aïr splits the problem in three: an ear, a brain and a mouth, running as one live chain.
The ear
Turns Hausa speech into something a language model can read. Today that is still a transcript.
The brain
A third-party multilingual model, McGill’s AfriqueQwen, decides what to say. We do not retrain its weights.
The mouth
Built on Kyutai’s Moshi architecture, it speaks while it keeps listening.
- Done · Oct 2026Thinker–talker prototype runningEar, brain and mouth run as one live research chain. Internal; write-up to follow.
- In progressAïr speech-to-speech end to endTraining the ear on real phone audio so hearing no longer goes through a transcript.
- NextMeasure people firstRecord consented two-channel Hausa conversations in Niger and Nigeria; publish their timing statistics.
Two voices, one timeline Overlap, backchannels and interruptions, drawn the way the Lab will show every recorded conversation.
What we do not claim yet
- Aïr’s live demo is not end-to-end speech-to-speech today: its hearing still goes through a transcript.
- Aïr has not learned Hausa turn-taking. Its timing is a policy plus distributions from other languages until real Hausa conversation is measured.
- No Aïr result is published yet. Internal numbers will appear with their conditions, after review.
- The brain is a third-party multilingual model (McGill AfriqueQwen). We do not retrain its weights, and we did not train it.
Aïr is built in the open, one measurement at a time
The Hira programme publishes the questions, the progress and the results, including what does not work.