Gulf Arabic speech dataset
Gulf Arabic speech at volume, including genuine dual-channel call-centre audio, transcribed in dialect-retaining Arabic script and delivered PII-redacted.
| Language | Gulf Arabic |
|---|---|
| Endonym | خليجي |
| ISO 639-3 | afb |
| Speaker population | ~36 million |
| Primary regions | Saudi Arabia (east), UAE, Kuwait, Qatar, Bahrain, Oman |
| Dialects covered | Kuwaiti, Emirati, Qatari, Bahraini, Hasawi, Omani |
| Hours available / collectable | 1,500 hours available (incl. call-centre) · 2,000 hours/quarter collectable |
| Recording spec | 8 kHz telephony and 48 kHz studio, mono and dual-channel call recordings |
| Transcript availability | Arabic script with dialect-form retention; MSA normalisation optional |
| Consent basis | Two-party consent for call audio, with speaker notification logs retained |
| License terms | Perpetual, worldwide, non-exclusive; PII-redacted delivery standard |
| Related languages | Hausa, Latin American Spanish, Malaysian Malay, Zulu, Vietnamese |
Why Gulf Arabic data is hard to find
Almost all public Arabic data is Modern Standard Arabic, which nobody speaks conversationally, so ASR trained on it collapses on Gulf dialect. Dialect has no fixed orthography, meaning transcripts must follow a documented convention or they are inconsistent. Real call audio is legally sensitive and rarely licensable without a consent chain.
What we can collect
- —Dual-channel call-centre audio with two-party consent and PII redaction
- —Dialect-balanced read and conversational speech across six Gulf states
- —Code-switched Arabic–English business speech
- —Gulf Arabic–English translation pairs
Licensing and consent
Two-party consent for call audio, with speaker notification logs retained. Delivery is under perpetual, worldwide, non-exclusive; pii-redacted delivery standard. Call and conversational audio is redacted for personal information before hand-off, and every recording carries a documented consent chain you can audit.
Need Gulf Arabic data at scale?
Send us hours, dialect mix and domain — we'll come back with a collection plan.
Talk to our sourcing team