AI Models · Arabic Voice
18 DialectsEvery Arabic dialect,
natively understood.
Arabic is not one language, it is a family of 18 distinct spoken varieties across 22 countries. Hamsa builds separate STT and TTS models for each dialect, trained on real native-speaker data, not a normalization layer on top of Modern Standard Arabic.
13+
Arabic dialects supported
422M+
Native Arabic speakers covered
<15%
WER on colloquial Arabic
1 API
Switch dialect with one parameter
Try it
Hear the voices, dialect by dialect
Pick a dialect and press a voice to hear it.
Voices
No sign-upPick a dialect and listen to the voices in it.
Why it matters
A model trained on MSA doesn't understand what your users actually say
Modern Standard Arabic is a formal written register used in news and education. No one grows up speaking it. The 422 million people you need to reach speak Egyptian Masri, Gulf Khaleeji, Levantine Shami, Moroccan Darija, and each of these differs from MSA as much as Italian differs from written Latin.
Generic Arabic STT models achieve 40–65% word error rate on real dialect speech. Hamsa trains native models for each dialect and reaches under 15% WER, the difference between a product that works and a product your users stop using after one sentence.
Measured on colloquial Arabic benchmarks · 2024
Browse by region
13+ Arabic dialects,
all supported natively
Gulf
Levant
Mesopotamia
Egypt & Sudan
Maghreb
Standard
One API, every dialect
Switch dialects with a single parameter
The same API endpoint handles every supported dialect. Pass dialect: "egyptian" for Cairo,dialect: "saudi" for Riyadh, or dialect: "auto"to let Hamsa detect and route automatically.
// Egyptian call center
hamsa.stt.stream({ dialect: 'egyptian' });
// Saudi IVR
hamsa.stt.stream({ dialect: 'saudi' });
// Let Hamsa detect automatically
hamsa.stt.stream({ dialect: 'auto' });


