Sada
Arabic speech recognition trained for meeting audio, and the model the Sada meeting intelligence platform runs on.
Sada is Misraj AI's Arabic speech recognition model, and the first stage of the Sada meeting intelligence platform. It takes the meeting audio and produces the transcript that the platform's analysis agents then read.
What separates it is ownership. Competing meeting tools license a speech model and adapt it to Arabic. This one was trained for Arabic, and it belongs to Misraj AI, so when it mishears a speaker the model is ours to correct rather than a support ticket filed with a vendor.
It runs where the audio already is. Meeting audio is among the most sensitive content an organization produces, so the model deploys on-premises and into a sovereign environment as well as through the cloud API, and it is available on its own rather than only inside the platform.
There is no published word error rate for Arabic meeting speech yet. No good public benchmark exists for multi-speaker, dialectal, code-switched Arabic recorded in real conditions. Until one does, this page is carried by what the model is and where it runs, and no number is estimated in the meantime.
Capabilities
- Arabic is the training target, not an adaptation. Trained for Arabic rather than licensed as a general speech model and tuned onto it afterwards.
- Meeting audio, not read speech. In production inside the Sada platform, against the conditions meetings actually have: several speakers, dialect, code-switching mid-sentence, and a real room.
- On-premises, sovereign, or cloud API. The same model in all three. For meeting audio, on-premises is the first option rather than the enterprise upgrade.
- Available on its own. Sold and callable separately from the Sada platform, not only as a component inside it.
- No published benchmark yet. Word error rate on Arabic meeting speech is not published, because the benchmark it would be measured against does not exist. A figure borrowed from an easier benchmark would tell you nothing about your meetings.
Access
Let's talk about what you're trying to build.
Tell us the problem. We'll tell you honestly whether AI is the right answer.