About
Software should understand people the way they speak
Nabra turns natural speech into reliable structured data, so businesses can accept requests by voice without building speech AI themselves.
The problem
A large share of business requests still start as speech — a phone call to a delivery company, a voice note to a shop, a technician describing a job. Someone then listens and re-types it into a system. It is slow, it loses details, and it doesn't scale.
Why voice
Speaking is faster than typing on a phone, works for people who don't like forms, and is already how many customers in our region prefer to communicate. Software shouldn't force them to switch.
The technology
Speech recognition turns audio into text; a language model reads the meaning and maps it to fields; a deterministic validation layer checks every value against what was actually said. The pieces are replaceable — the API contract is the product.
Arabic dialects and English
Most voice tools are tuned for English and treat Arabic as one language. People speak Levantine, Gulf, English — often mixed in one sentence. Nabra detects which, understands it, and asks follow-up questions in the user's language — clear, formal Arabic for Arabic speakers.
Developer first
One HTTP call, documented behaviour, explicit statuses and confidence for every field. Your application keeps its business logic; we don't create orders or touch your systems.
Privacy first
Audio is processed and deleted; recordings aren't kept unless you explicitly enable debugging. The service is built to run on your own infrastructure — that's how it runs today.
Where we are
Nabra is at an early stage. The Voice API works end to end as a self-hosted prototype — speech recognition, structured extraction and validation, tested on English, ten Arabic dialects and mixed speech (text), plus real Jordanian recordings. A hosted service, accounts and billing are on the roadmap. We would rather show you what works than promise what doesn't yet.
Define your data. Let people speak.
Try the live demo, or build a schema and test it with your own voice in a few minutes.