Arabic Dialects · Speech to Text

MSA · فصحى

Modern Standard Arabic
Speech to Text

An MSA STT model that handles the wide spectrum of regional accents as speakers from different dialect backgrounds produce formal Arabic in media, education, courts, and official settings.

dialect: "msa"MENA region · real-time · multiple regional accents

420M+

Modern Standard Arabic users

<14%

WER on spoken Modern Standard Arabic

<300ms

First-word streaming latency

MENA

Region-wide official language coverage

Try it

Speak Modern Standard Arabic.
Watch it transcribe.

Record yourself speaking Modern Standard Arabic and watch the words appear as you talk. The demo runs Hamsa's full production STT, the same model your API calls will use.

Speech to Text · speech in → text out

No sign-up
Microphone
press to record

Your words will appear here.

Press record and speak Arabic or English. The words appear as you talk.

Free to start · in your browserTranscribe in Hamsa Media

Register overview

What is Modern Standard Arabic?

Modern Standard Arabic, known as al-Fusha (الفصحى), is the unified, standardized form of Arabic used in media, education, law and official contexts across all Arab countries. It is nobody's mother tongue. It is a learned register, consciously produced by broadcasters, judges, teachers and politicians throughout the Arab world, so every speaker carries traces of their first dialect into its sounds and stress.

Key features that shape spoken MSA:

  • A learned register, not a mother tongue: acquired at school and used on formal occasions.
  • The jim shifts with the speaker's background: /ʒ/ in the Levant and the Maghreb, /g/ for many Egyptian speakers.
  • Case endings are voiced in formal reading, while word endings are usually dropped at a pause.
  • Modern Arabized terms: "الحوسبة السحابية" (cloud computing), "الذكاء الاصطناعي" (artificial intelligence), "التحول الرقمي" (digital transformation).

Reach

420M+

Across the MENA region

Primary context

4

Media · Education · Law · Official

Unique property

Learned

A learned register, not a native language

Hamsa WER

<14%

on spoken MSA · 2024

The problem

Why generic Arabic STT stumbles
on spoken MSA

MSA is formal, but the people speaking it are not uniform. Here is why.

Accent diversity misleads generic models

Speakers from 22 countries pronounce MSA with different accents. A model trained on one accent fails with the others.

Fine differences between Quranic recitation and media MSA

Pronunciation, pace and rhythm differ between Quranic recitation, media MSA and political speech.

Technical and newly coined terms

Modern Arabized technical and scientific terms can trip up MSA models that have not been kept up to date.

Rhythm that varies by speaker and context

A news anchor speaks at a completely different rhythm from a judge, a university professor or a politician.

The solution

How Hamsa understands spoken MSA
in every accent

✓

Training on MSA from many regional backgrounds

Training data covering MSA speakers from the Maghreb, the Gulf, the Levant, Egypt and Iraq.

✓

Coverage of diverse usage contexts

Training on media, education, law and political speech, each with its own distinct style.

✓

An up-to-date terminology lexicon

Support for modern technical, scientific and media terminology.

✓

A model that handles varied rhythm

From fast-paced speeches to slow, deliberate reading.

Capabilities

Everything you need for
Modern Standard Arabic STT

✓

Regional Accent Coverage

One model that understands MSA as spoken by speakers from the Maghreb, the Gulf, the Levant, Egypt and Iraq, with no separate setup for each accent.

✓

Speaker Diarization

Identify and label individual speakers in interviews, court sessions, and multi-party conferences.

✓

Smart Punctuation

Sentence boundaries, commas, and question marks placed according to the rules of formal written Arabic, for text that is ready to publish and archive.

✓

Real-time Streaming

Under 300ms first-word latency. Transcript fragments arrive as the speaker talks, suited to live broadcasts and real-time captioning.

✓

Custom Vocabulary

Add legal, scientific and technical terms, proper names, and institution names to improve recognition accuracy in your domain.

✓

Word-level Timestamps

Precise timing for every word makes it easy to caption programmes and index media and educational archives.

Transcript examples

Formal Arabic.
Accurate transcription.

News bulletin

أعلنت وزارة الصحة اليوم إطلاق حملة وطنية للتطعيم تستمر أسبوعين في جميع المحافظات.

"The Ministry of Health today announced the launch of a two-week national vaccination campaign across all governorates."

Broadcast register: verb-first sentence, standard news vocabulary

University lecture

ننتقل الآن إلى المحور الثاني من المحاضرة، وهو أثر الذكاء الاصطناعي في التحول الرقمي.

"We now move on to the second part of the lecture: the impact of artificial intelligence on digital transformation."

Modern Arabized terms: "الذكاء الاصطناعي" (AI), "التحول الرقمي" (digital transformation)

Court session

قررت المحكمة تأجيل النظر في الدعوى إلى جلسة الأسبوع المقبل لسماع أقوال الشهود.

"The court decided to adjourn the case to next week's session to hear the witnesses' testimony."

Legal register: "الدعوى" (the case), "أقوال الشهود" (witness testimony)

Use cases

Where MSA STT
delivers results

Broadcast Media

News and programme transcription

Converting news bulletins, programmes, and interviews to accurate text for archiving and translation.

E-Learning

Lecture and course transcription

Converting university lectures and formal training courses to searchable text.

Courts & Justice

Court session documentation

Recording and transcribing court proceedings in formal MSA.

Speeches & Conferences

Political speech and conference transcription

Converting official speeches and international conferences to accurate text documents.

Accuracy benchmark

High-accuracy STT for Modern Standard Arabic

Measured on spoken MSA from speakers with many regional backgrounds, in news bulletins, lectures and official speeches. Lower word error rate means more words transcribed correctly.

Word Error Rate (WER) · lower is better

Hamsa STT (MSA)us14%
Generic Arabic STT38.4%
Single-accent model48.1%
GPT-4o-transcribe52.6%

Evaluated on spoken MSA across multiple regional accents · 2024 · Lower is better

Architecture

How it works

01

Audio in

Send audio via WebSocket stream or REST upload. Accepts PCM, MP3, M4A, WAV, and WebM, so it works with broadcast studios, courtroom and conference recording systems, and education platforms.

02

Accent and context recognition

Hamsa recognizes the speaker's regional accent and pace, from fast-paced speeches to slow reading, and applies the MSA acoustic and language models together with the modern terminology lexicon to keep the transcript accurate.

03

Transcript out

Transcripts stream as JSON events with word-level timestamps, speaker labels, confidence scores, and punctuation. Pipe directly into your archive, captioning system, or downstream NLP pipeline.

Developer API

Two lines to Modern Standard Arabic speech recognition

Drop in the Hamsa SDK, set dialect: "msa", and start streaming transcripts. No preprocessing pipeline, no separate setup for each accent.

Read the API docs →
msa-stt.tsstreaming
import { Hamsa } from '@hamsa-ai/sdk';

const stream = hamsa.stt.stream({
  model:       'hamsa-stt-s4',
  dialect:     'msa',
  diarization: true,
});

stream.on('transcript', (t) => console.log(t.text));
// → "تُعقد الجلسة الافتتاحية للمؤتمر صباح الغد." · 00:01.840

Related services

More Hamsa for Modern Standard Arabic

Why Hamsa

Built for Arabic from the ground up

✓

Native, not adapted

Hamsa trains on real spoken Arabic, not English models adapted for Arabic. Its MSA model learns from speakers across the Arab world, not a single studio accent.

✓

Enterprise-grade SLA

99.9% uptime, SOC 2 Type II compliance, and data residency in the MENA region. Your audio stays in-region.

✓

Fastest Arabic streaming

Under 300ms first-word latency in production. Real-time transcription that keeps pace with live broadcasts and speeches.

✓

13+ dialects, one API

Switch between MSA and any dialect with a single parameter. Transcribe the newscast and the call-in segment with the same integration.

FAQ

MSA STT,
frequently asked

What is the difference between Modern Standard Arabic and the Arabic dialects?

Modern Standard Arabic is the unified, standardized form of Arabic used in media, education, law and official contexts across all Arab countries. Nobody speaks it as a native language at home. It is acquired through learning and consciously produced on formal occasions. The dialects are the languages of everyday conversation, each with its own pronunciation and vocabulary.

Does the model handle speakers from different dialect backgrounds?

Yes. Every speaker pronounces MSA with a dialect background that shapes their sounds and stress. Hamsa's model is trained on spoken MSA from speakers in the Maghreb, the Gulf, the Levant, Egypt and Iraq, so it handles this diversity with a single model.

What word error rate does Hamsa achieve on MSA?

Hamsa achieves a word error rate (WER) under 14% on spoken Modern Standard Arabic, across varied contexts including news bulletins, lectures and official speeches.

Is the model suited to documenting court sessions and conferences?

Yes. The model supports speaker diarization, word-level timestamps and punctuation that follows MSA rules, so it can produce accurate written records of court hearings, conferences and official speeches.

When should I use the MSA model and when a dialect model?

Use the MSA model for prepared formal speech such as news bulletins, lectures, court hearings and speeches. For everyday conversations and customer service calls, the matching dialect model is more accurate, because speakers use their own dialect rather than MSA.

Explore more

13+ Arabic dialects, supported

Partner ecosystem

Trusted Across the
Technology Ecosystem

Hamsa works with leading technology companies, integrators, and solution providers to bring enterprise Voice AI to organizations across the region.