Arabic voice AI — built dialect-first.
We build real-time voice agents for Arabic-speaking markets, and the speech infrastructure underneath them. Most Arabic speech tooling is trained on Modern Standard Arabic and falls apart on how people actually talk — so we build our own: dialectal corpora, text-to-speech voices, and turn-taking models tuned for Egyptian and Gulf Arabic.
| Area | What it means in practice |
|---|---|
| Text-to-speech | Natural dialectal Arabic voices — Egyptian and Gulf — for conversational agents |
| Speech recognition | ASR adapted to dialectal speech, including telephony-grade audio |
| Turn detection | End-of-utterance models so agents know when a caller has actually finished |
| Voice agents | Low-latency real-time pipelines for phone and web |
Speech corpora are the bottleneck for dialectal Arabic, so we build our own — collected, segmented, transcribed and quality-checked in-house. Datasets published on this profile are released for research and evaluation; some are private and available on request.
Current releases are listed below.
We take on voice-agent and Arabic speech projects, and we license our dialectal corpora and voices. Get in touch:
Interested in a dataset that isn't public? Open a discussion on this profile or email us.