Home/Articles

Articles

Field notes on Arabic speech data, dialect coverage and evaluating voice AI for MENA.

Latest

19 October 20263 min read

Najdi vs Hijazi vs Gulf Arabic: Why Your ASR Model Fails on Saudi Dialects

Today’s best Arabic ASR models get 36–48% of words wrong on Saudi dialects, compared with 19% on MSA. Learn why Najdi, Hijazi and Gulf Arabic break speech models, and what kind of training data closes the gap.

Read article

Building voice AI for Arabic speakers?

Tell us the dialects and use case. We'll scope a pilot dataset with you.

Request a Pilot