Arabic Dialects · Speech to Text

Moroccan · Darija

Moroccan Darija
Speech to Text

Hamsa STT understands Moroccan Darija the way Moroccans actually speak it, with its natural blend of Arabic, Amazigh, and French. Trained on real Moroccan speech for BPO, telecom, financial services, and e-commerce.

dialect: "moroccan"35M+ speakers · real-time · quadrilingual

35M+

Moroccan Darija speakers

<31%

WER on Moroccan Darija speech

<300ms

First-word streaming latency

Quadrilingual

Arabic + Amazigh + French + Spanish

Coming Soon

Demo launching soon

The interactive demo for this dialect is coming soon. Sign up to be notified when it launches.

Join the waitlist →

Dialect overview

What is Moroccan Darija?

Moroccan Darija (الدارجة) is the everyday spoken language of more than 35 million Moroccans and of large diaspora communities in Europe. Arabic is its foundation, Amazigh its historical weave, French present in professional and technical vocabulary, and Spanish audible in northern Morocco. It varies between Casablanca, Rabat, Fes, Marrakech, and the North, which makes it one of the most complex varieties in the Arab world.

Key features that distinguish Darija from Modern Standard Arabic:

  • Negation with "ما...ش": "ما فهمتش" means "I did not understand".
  • The "كا" or "تا" prefix for the present tense: "كنهضر" (I am talking), "تانمشي" (I am going).
  • French words woven into daily speech: "الطوموبيل" (car), "البورطابل" (mobile phone), "ديريكت" (directly).
  • Common Darija expressions: "بزاف" (a lot), "واخا" (OK), "دابا" (now), "واش" (question marker).

Native speakers

35M+

Morocco

Regional varieties

Multiple

Casablanca · Fes · Marrakech · North

BPO standing

#1

Europe's top BPO destination

Hamsa WER

<31%

on real Moroccan Darija speech · 2024

The problem

Why generic Arabic STT fails
on Moroccan Darija

Darija is fast, multilingual, and far from the MSA most Arabic models are trained on. Here is why they break.

MSA models do not understand Darija

Moroccan Darija sits far from MSA in pronunciation, vocabulary, and grammar, with dropped short vowels, dense consonant clusters, and everyday forms like "كنهضر" (I am talking) and "ما فهمتش" (I did not understand). Models trained on news bulletins and read text treat this speech as noise or unknown words, and produce a transcript that does not reflect what the speaker said.

Quadrilingual speech confuses generic models

Moroccans blend Arabic, Amazigh and French, and Spanish in the north, within a single sentence without feeling they are mixing languages. Generic models assume one language per segment, so they drop French words or spell them wrongly, which directly hurts transcription accuracy.

City and regional variation

Darija differs between Casablanca, Rabat, Fes, Marrakech, and the North in pronunciation, rhythm, and vocabulary. A model that ignores these differences fails a large share of speakers and treats local variation as speech errors.

BPO calls need accuracy and speed together

Moroccan BPO centers serving European clients handle Darija and French calls at scale. Any transcription delay or drop in accuracy raises the cost of quality monitoring and slows daily operations.

The solution

How Hamsa handles
Moroccan Darija natively

✓

Authentic Moroccan training data

Hamsa trains its model on real Moroccan speech from contact center calls and everyday conversation, not repurposed MSA and not synthetic audio generated with text-to-speech.

✓

Quadrilingual recognition

The model automatically detects Arabic, Amazigh, French, and Spanish segments within the same conversation, and processes each in its own language so every word is written correctly.

✓

City varieties in one model

Acoustic models cover speech patterns from Casablanca, Fes, Marrakech, and the North, routing each utterance automatically to the best-matched model, with no separate setup per region.

✓

Sub-300ms real-time streaming

Infrastructure optimized for contact centers and BPO, where text arrives while the speaker is still talking, so quality monitoring and agent assist work in real time.

Capabilities

Everything you need for
Moroccan Darija STT

✓

Native Darija Transcription

Trained on colloquial Moroccan speech as it is actually spoken. Not MSA with a dialect layer on top, but native Moroccan acoustic and language models from the ground up.

✓

Code-Switch Detection

Moroccan speakers move between Arabic, Amazigh and French, and Spanish in the north, within the same sentence. Hamsa handles these transitions without losing context or misidentifying the language.

✓

Speaker Diarization

Identify and label individual speakers in call recordings. Critical for Moroccan BPO centers processing thousands of calls every day.

✓

Smart Punctuation

Sentence boundaries, commas, and question marks placed according to Moroccan Darija structure, not MSA grammar rules.

✓

Real-time Streaming

Under 300ms first-word latency. Transcript fragments arrive as the speaker talks, with no silence-detection delay and no waiting for utterance boundaries.

✓

Custom Vocabulary

Add Moroccan brand names, banking and technical terms, and your industry jargon to improve recognition accuracy for your domain.

Transcript examples

Real Moroccan Darija.
Accurate transcription.

Call center (BPO)

واخا، غير تسنا شوية، دابا غادي نشوف ليك الكونط ديالك.

"OK, just wait a moment, I'm going to check your account for you now."

Darija: "واخا" (OK), "دابا" (now), "غادي" (going to), "الكونط" (account, from French compte)

Customer service

راه عيطت ليكم البارح وقالو ليا غادي يتحل المشكل، ولكن مازال ما تقاد والو.

"I called you yesterday and they told me the problem would be fixed, but nothing has been sorted yet."

Natural Darija: "عيطت" (I called), "البارح" (yesterday), "ما تقاد والو" (nothing got fixed)

Banking call

بغيت نبدل l'adresse ديالي، حيت la carte ما وصلاتنيش.

"I want to change my address, because the card never reached me."

Code-switching: French "l'adresse" and "la carte" mid-sentence, "ما...ش" negation

Use cases

Where Moroccan Darija STT
delivers results

European BPO

French/Spanish BPO call automation

Transcribing and analyzing calls at Morocco's BPO centers serving European clients in Darija and French.

Telecom

Customer service call transcription

High-accuracy transcription for Moroccan telecom company customer service.

Financial Services

Banking call documentation

Transcribing banking and financial service calls for compliance and quality management.

E-Commerce

Customer call transcription

Accurate recognition of Moroccan e-commerce customer queries.

Accuracy benchmark

High-accuracy STT for Moroccan Darija

Measured on a held-out benchmark of real Moroccan call center and conversational audio. Lower word error rate means more words transcribed correctly.

Word Error Rate (WER) · lower is better

Hamsa STT (Moroccan)us31%
Generic Arabic STT38.4%
MSA-trained model48.1%
GPT-4o-transcribe52.6%

Evaluated on colloquial Moroccan Darija speech benchmarks · 2024 · Lower is better

Architecture

How it works

01

Audio in

Send audio via WebSocket stream or REST upload. Hamsa accepts PCM, MP3, M4A, WAV, and WebM, so it works with any microphone, telephony platform, or BPO recording system.

02

Dialect routing

Hamsa detects Moroccan Darija features and routes audio to the Darija-specific acoustic and language models. When code-switching is detected, Arabic, Amazigh, French, and Spanish segments are processed in parallel to keep transcription accurate and context intact.

03

Transcript out

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

Developer API

Two lines to Moroccan Darija speech recognition

Drop in the Hamsa SDK, set dialect: "moroccan", and start streaming transcripts. No preprocessing pipeline, no normalization to MSA, no fallback surprises.

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

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

stream.on('transcript', (t) => console.log(t.text));
// → "واش كاين شي مشكل فالكونط ديالك؟" · 00:01.840

Related services

More Hamsa for Moroccan Darija

Why Hamsa

Built for Arabic from the ground up

✓

Native, not adapted

Hamsa trains on real Arabic dialect data, not English models adapted for Arabic. The difference is 25 to 50 WER points on colloquial speech.

✓

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 human speech rates.

✓

13+ dialects, one API

Switch dialect with a single parameter. Run A/B tests across Moroccan Darija and the other Maghrebi dialects. No separate integrations, no separate contracts.

FAQ

Moroccan Darija STT,
frequently asked

What is the difference between Moroccan Darija and Modern Standard Arabic?

Modern Standard Arabic is the formal register used in media, education, and official documents. Darija is the everyday spoken language of more than 35 million Moroccans. It differs from MSA in pronunciation, vocabulary, and grammar, and blends Arabic with Amazigh, French, and at times Spanish. That is why STT models trained on MSA struggle to understand Darija as it is actually spoken.

Does Hamsa handle Darija and French mixed in the same call?

Yes. Switching between Darija and French within a single sentence is normal in Moroccan speech, especially in professional calls and BPO centers. Hamsa detects these switches in real time and processes each segment in its own language. It also handles the Amazigh and Spanish vocabulary present in everyday speech.

What word error rate does Hamsa achieve on Moroccan Darija?

Hamsa achieves under 31% WER on real Moroccan Darija speech. The gap with generic models is widest on informal speech, such as contact center recordings and natural, unscripted conversation.

Does Hamsa understand the different Moroccan city varieties?

Yes. Darija varies between Casablanca, Rabat, Fes, Marrakech, and the North in pronunciation, rhythm, and vocabulary. Hamsa covers these patterns in one model and automatically routes each utterance to the best-matched model for its features.

Is Moroccan Darija STT suitable for BPO centers?

Yes. The model is optimized for European BPO and contact centers, with sub-300ms real-time streaming, speaker diarization, and support for Darija and French in the same call, enabling transcription and quality monitoring at scale.

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.