Arabic Speech Recognition Datasets Compared: Dialects, Hours & Licences

The short answer: the largest public Arabic speech recognition datasets are QASR (about 2,000 hours) and MGB-2 (about 1,200 hours) of Al Jazeera broadcasts, followed by MASC (1,000 hours of YouTube audio) and SADA (about 667 hours of Saudi television). Common Voice, FLEURS, Casablanca, Mixat and a few telephone corpora from the LDC fill smaller niches. Most of these hours are broadcast or read speech, and much of the audio is Modern Standard Arabic (MSA). Several of the biggest sets are licensed for non-commercial use only. Conversational Gulf speech, contact-centre telephony and code-switched speech are the thinnest slices, and they are usually what a production system hears.

Arabic speech recognition datasets at a glance

"Hours" is the figure the source states; some corpora count raw audio and others count transcribed segments, so treat small differences as noise. "Light" transcription means existing captions or transcripts were aligned to audio automatically. "Full" means humans transcribed the audio segment by segment.

Need commercial rights? Bayanat Voice is our #1 pick for licensed Arabic speech

The open corpora above are the right place to start research and baselines. When your Arabic speech recognition model has to ship in a product, you need audio from the channels your users actually speak on and a licence that permits it. That is where Bayanat Labs is the best partner, and why Bayanat Voice is our #1 pick for commercially licensed Arabic and Saudi speech:

• Saudi speech at production volume. Saudi Spontaneous Speech is 1,000 hours of unscripted Saudi speech with task annotation and commercial AI-use rights. Saudi Conversational Speech is a separate 1,000-hour collection with human transcripts, timestamps, speaker attributes and a fixed delivery schema.
• Customer-service and call-center audio. Saudi Full-Duplex Customer-Service Speech gives you 350 hours of two-way service audio with task annotation and commercial AI-use rights. Arabic Call-Center Conversations and Arabic Scripted Monologues together make up 10,000 hours, sold as separate products; ask for the allocation to each.
• Recent smartphone recordings. Arabic Smartphone Speech is 500 hours collected in 2025–2026, with scripted and spontaneous subsets labelled separately.
• Rights you can read in the agreement. Permitted commercial AI use is written into the licence, not inferred from a dataset card. Training, retrieval/display, redistribution and synthetic-voice scopes are kept separate, so you know exactly what you can ship.
• One supplier for licensed and custom data. If your dialect or channel is missing, the same team collects and transcribes it with vetted native speakers across 25+ Arabic varieties who pass dialect screening, calibrated against gold standards with adjudication and an audit trail on every task.
• Sovereign, model-agnostic, fast to test. Hosting is in-region by default, with on-prem and private-cloud options. We do data only and never compete with your model. A fixed-scope pilot with a quality report usually lands within two weeks.

To see how these collections line up against every other commercial source, from vendor catalogues to LDC and ELRA licences, read our commercial Arabic datasets comparison.

Broadcast and MSA corpora

The big Arabic ASR datasets come from television news. MGB-2 released about 1,200 hours from 19 Al Jazeera programmes aired between 2005 and 2015, and its authors estimate that more than 70% of the speech is MSA. QASR extends the same archive to about 2,000 hours, with multi-dialect and code-switched segments and linguistically motivated segmentation. Both are lightly supervised: existing human transcripts, which can omit or correct what speakers actually said, were aligned to the audio automatically. The Casablanca paper flags this as a source of mismatch between transcript and speech.

SADA is the Gulf counterweight. It comes from Saudi television rather than a pan-Arab news channel, and its DataPort listing labels each segment's dialect, speaker type and recording environment. That makes it the most useful public set for Saudi varieties, as long as your use is non-commercial. MASC adds 1,000 hours crawled from more than 700 YouTube channels under a CC BY 4.0 card. Its card calls it multi-regional, multi-genre and multi-dialect; check its per-clip metadata before relying on it for any one variety.

Broadcast audio suits pre-training, media monitoring and captioning. It is a poor proxy for a customer calling a bank from a car: presenters speak clearly, one at a time, into studio microphones, and lean toward MSA.

Read and crowd-sourced speech

Read-speech sets give clean pairs of audio and known text. The Common Voice Scripted Speech 27.0 metadata lists 158 recorded and 93 validated hours of Arabic from 1,679 voices (collection cut-off 11 September 2026), released as CC0-1.0. FLEURS reads translated FLoRes sentences, with around 10 hours of training data per language. Its own card warns that read-speech results can differ from performance in noisier, real-world settings. The Arabic Speech Corpus is a single-speaker studio set built for speech synthesis, not recognition; our comparison of Arabic speech synthesis datasets covers it and the other open TTS corpora.

These sets suit smoke tests and licence-clean fine-tuning. None gives you curated dialect labels, and people do not talk to a voice assistant the way they read a sentence off a screen.

Dialectal and conversational corpora

This is where public data gets thin. Casablanca describes itself as the largest fully supervised Arabic dialect dataset: about 48 hours of human transcription across eight countries, with gender, dialect and English/French code-switching labels. It ships YouTube URLs rather than audio, because of copyright, so clips can disappear. Its authors also note that more than 60% of speakers are male in every dialect except Moroccan. MGB-3 (16 hours of Egyptian YouTube) and MGB-5 (about 13 transcribed hours of Moroccan) are smaller still.

Conversational code-switching has one notable public set: Mixat, 15 hours of Emirati speech mixed with English from two podcasts. Its authors built it because current speech recognition resources fall short on Emirati speech. Telephone speech lives mostly in the LDC catalogue: Gulf Arabic Conversational Telephone Speech (about 2,800 minutes from 975 speakers) and CALLHOME Egyptian (120 calls). CALLHOME was published in 1997 and the Gulf corpus was recorded in 2004; both are casual calls, not service interactions.

Dialect labels are another weak point. The Arab Voices project (January 2026), which harmonises 31 datasets across 14 dialects, reports that corpora define dialect by country, by coarse region or by ISO-style codes, which makes pooling them unreliable. It also notes that metadata such as speaker demographics is often limited or absent. Automatic dialect labels will not rescue you: in LabelBench, a local language model reached 22.4% exact-match accuracy across five dialect regions from text, on the fixed held-out manifest (chance is 20%). For more on building dialect labels that hold up, see Arabic dialect identification.

Here is why the style gap matters. The same request, "I want to check my balance", looks like this across the registers a banking voicebot will hear:

A model trained mostly on the first row has seen the noun and little else. The verb of wanting, the verb of looking and the English insert change from row to row, and each also has several spellings (أبغى / ابغى / أبغا). Our guide to Arabizi and code-switching covers how to label the last row consistently.

Licences: what you can use commercially

Read the licence on the card, then the provenance. The licence says what the curators permit; provenance says whether they had the rights to permit it.

Three traps catch teams most often:

• Web-sourced audio. A CC BY card on a corpus crawled from YouTube or TV covers the curators' work. It does not transfer the broadcaster's or the uploader's rights in the audio. Casablanca's decision to ship URLs instead of files shows its authors took this seriously.
• Models trained on non-commercial data. Whether weights trained on NC data can be used commercially is a legal question the licence text does not settle. Get advice before a model trained on SADA or QASR ships in a product.
• Voices are personal data. Speech can identify a speaker, and your own call recordings fall under privacy law. See our PDPL data residency checklist before customer audio leaves your systems for transcription.

What is missing from public Arabic speech data

Map your deployment against this list. Every item you need is a gap public data will not close.

1. Contact-centre telephony. None of the corpora above covers 8 kHz banking, telecom or government service calls, with hold music, cross-talk and account numbers read aloud.
2. Conversational Saudi and Gulf speech on your channel. SADA labels Najdi, Hijazi and Khaliji, but it is television. The LDC Gulf telephone corpus was recorded in 2004. For contact-centre audio, our Arabic telephony speech data comparison covers the vendor sets, 8 kHz specs and PDPL. For the varieties themselves, see Saudi Arabic dialects for AI.
3. Code-switching beyond one pairing. Mixat covers Emirati with English. Gulf English in Riyadh offices, Maghrebi French and Levantine English inserts have almost nothing public at conversational scale.
4. Smaller varieties. Casablanca's authors single out Emirati, Yemeni and Mauritanian as under-represented, yet Casablanca releases only validation and test splits, and MGB-5's transcribed Moroccan set is about 13 hours.
5. Balanced demographics. Gender, age band and region are recorded unevenly across corpora, and some sets skew heavily male.
6. Domain vocabulary. Clinical dictation, legal proceedings and financial terms read in dialect are absent from public sets.
7. Consistent spelling. Dialects have no official orthography. In MGB-3, four transcribers still disagreed on about 13% of words after normalising common letter variants. Mixing corpora mixes their conventions, which adds noise to both training labels and WER.

The effect shows up in evaluation. On SADA, the Open Universal Arabic ASR Leaderboard (December 2024 paper) found that every model tested did best on MSA and declined significantly on Egyptian and Khaliji. The authors attribute this to how heavily public training data leans toward MSA. Our piece on why dialect coverage sets your model's ceiling covers the same effect in text models.

Several of these gaps are exactly what Bayanat Voice was built to close: Saudi full-duplex customer-service speech, call-center conversations, Saudi spontaneous and conversational speech, and recent smartphone recordings, all under a commercial licence. Start there, compare recording type, included annotations and permitted use against your deployment, and commission custom collection only for what remains.

How to choose Arabic speech data by recording type and task

Hours only tell you how much audio there is, not whether it fits your model. A thousand hours of read prompts and a thousand hours of real service calls teach a model different things. Choose by two questions first: how was the audio produced, and what will the model do with it. Then compare hours within the rows that fit.

Most public corpora above are broadcast or web video, which sits outside these rows, or read speech. Production systems hear the first three rows. That is also why one large total, across collections of different kinds, says little: ask for the hours in each row you need.

Eight things to check in a sample before you license

Ask for the datasheet and a sample, then check each field below against your deployment. A red flag does not always rule a collection out, but it should be answered in writing before you sign.

The last row matters more than it looks. When the same speaker, or two halves of the same call, land in both training and test data, word error rate can look better than it will be in production, because the model has already heard that voice and that conversation. Split by speaker and by session, and keep a held-out set from your own channel that no supplier has seen.

Which Bayanat Voice collection fits which task

Bayanat Labs sells each speech collection as its own product, so you can pick by recording type instead of buying a single total. Here is how the six Bayanat Voice collections map to the tasks above. Facts not yet published are listed under "Ask for" rather than estimated.

A common combination is scripted or smartphone audio for acoustic coverage plus conversational or full-duplex audio for the way people really talk, with a held-out slice of your own calls for testing. Every collection comes under a commercial licence. Permitted training use is set in the agreement, retrieval or display and redistribution are separate scopes to name in your request, and synthetic-voice use is not implied by recognition rights. When a row you need is missing, for example a channel, dialect or domain, our speech data service records and transcribes it to a written specification. Request a speech datasheet and sample for the collection and task you have in mind.

Speech recognition, prompt testing or a synthetic voice: which Arabic speech data fits?

Everything above is about teaching a model to recognise speech. Building a voice is a different purchase. A recognition corpus wants many speakers, real channels and messy speech. A text-to-speech (TTS) voice wants the opposite: a few consistent speakers, clean studio audio, carefully prepared text and written permission to reproduce the voice. Between the two sits a smaller job, controlled prompt testing, where a fixed list of prompts read by known speakers shows how a system handles them. Decide which of the three you are buying before you compare hours.

Most of the corpora on this page sit in the first column. The open corpora built for synthesis, with their hours, diacritization and consent records, are compared in our guide to Arabic speech synthesis datasets.

What the Bayanat Voice Studio collections contain

Studio Prompt Speech is 2,000 Arabic prompts, 40 speakers and 2 hours of audio, sold for controlled speech testing and voice prototyping. The three numbers describe the product as a whole: they do not mean every speaker read every prompt, so ask how prompts map to recordings and speakers. Two hours in total is small for voice training whatever the speaker split, so use it as stated, for testing and prototypes, not as a voice-training corpus.

Saudi Expressive Speech is recorded by professional voice actors, with task annotation and commercial AI-use rights. Its collection size, speaker count and style breakdown are available on request rather than published. Synthetic voice creation, reuse of a speaker's voice and distribution need explicit contractual scope; the commercial licence does not imply them. When neither collection fits, custom studio recording covers single- or multi-speaker voices, pronunciation coverage, lexicons, expressive styles and brand-voice briefs.

Seven speaker-rights questions to answer before you build a voice

Use these as an intake sheet when you request any studio or voice data, from Bayanat Labs or anyone else. Each answer should be in the agreement, not in an email. This is not legal advice; your counsel decides whether the terms fit your product.

Why Bayanat Voice Studio is our #1 pick for Arabic voice data

For teams testing or building an Arabic voice, Bayanat Voice Studio is our #1 pick, because voice permissions are agreed in the contract before you record:

• Three clear buying paths. A compact studio prompt collection (2,000 prompts, 40 speakers, 2 hours) for controlled testing, Saudi expressive speech by professional voice actors, and custom studio production when you need your own voice.
• Voice permissions written into the contract. Synthetic voice creation needs explicit terms, and brand-voice agreements define permitted users, exclusivity and reuse.
• Rights kept separate. Training, retrieval/display, redistribution and synthetic-voice scopes are named separately, so recognition rights are not read as voice rights.
• Native Arabic production. Custom recordings draw on Bayanat's vetted native speakers across 25+ Arabic varieties, who pass dialect screening; for custom work, scripts, pronunciation and style labels are reviewed before production.
• Listen before you scale. A fixed-scope pilot with a quality report usually lands within two weeks, so you can hear the voice before the full script is recorded. Our speech data service runs the custom recordings.
• Sovereign by default. Hosting is in-region by default, with on-prem and private-cloud options, and we do data only, so we never compete with your model.

Request the Voice Studio specifications with your intended use: testing, prototyping or a voice you plan to ship.

When to commission custom Arabic speech data

Public corpora are the right start for pre-training, baselines and comparability. Commission custom collection or annotation when two or more of these are true:

1. Your deployment dialect is not a labelled variety in any corpus whose licence fits your use.
2. Your channel (phone, in-car, far-field, messaging voice notes) does not match studio or broadcast audio.
3. You need commercial rights and the matching corpus is NC, ND or has unstated terms.
4. Your users code-switch, and you need a written rule for how English or French inserts are transcribed.
5. You need domain terms, names or numbers that public sets never contain.
6. You need a held-out test set nobody has trained on, sliced by dialect, gender and age.

Even then, custom data rarely replaces public data. The usual pattern is to pre-train or start from a model built on broadcast hours, then fine-tune and evaluate on a smaller, well-specified in-dialect set. Write the transcription guidelines first, so the new hours share one convention. Set speaker quotas per variety and channel, not per "Arabic"; without them, a collection drifts toward whoever is easiest to record.

That second step is the work Bayanat Labs does. Our Arabic speech data collection and transcription is produced by vetted native speakers who pass dialect screening before they touch client audio. It covers transcription, timestamping, diarization, and accent and emotion tagging, calibrated against gold standards with adjudication and rework loops. Where you need broader dialect data collection across 25+ varieties, the same process applies. Hosting is in-region by default. A fixed-scope pilot with gold-standard QA and a quality report, usually within two weeks, lets you measure the data against your own test set before you scale.