Saudi Arabic conversational speech data
Free conversations in Najdi, Hejazi, Southern, Eastern and Northern Saudi Arabic: friends, family and colleagues talking about their day, plans and news, with the overlaps, laughter and English words of real speech. Every recording is transcribed word for word and checked by several native Saudi reviewers.
Specification
- Dialect
- Najdi, Hejazi, Southern, Eastern and Northern Saudi Arabic
- Conversations
- Two or more speakers talking freely: daily life, plans, family, work, sport and news
- Audio
- 16 kHz or 48 kHz, 16-bit WAV, recorded on phones and close microphones, indoors and outdoors
- Speakers
- Native Saudi speakers, balanced across gender, age groups and regions
- Annotation
- Verbatim transcript, timestamps, speaker turns, overlaps, laughter and noise tags; dialect and gender per speaker
- Quality check
- Every transcript is cross-checked by several native Saudi reviewers
Example of the data
- Speaker A00:01
وش رايك نطلع البر نهاية الأسبوع؟
How about going out to the desert this weekend?
- Speaker B00:04
والله فكرة، بس الجو حار هالأيام.
Good idea, but it’s hot these days.
- Speaker A00:09
خلاص نطلع العصر بعد ما يبرد شوي.
Then let’s go in the afternoon once it cools down a bit.
What it is used for
Dialect speech recognition
Teach ASR the way Saudi people really talk, not read speech.
Voice assistants
Understand casual requests, not only scripted commands.
Speaker diarisation
Separate who spoke when, through overlaps and interruptions.
Dialect identification
Tell Saudi Arabic apart from neighbouring dialects.
Questions
Are the conversations scripted?
No. Speakers talk freely about everyday topics, so the data has the hesitations, overlaps and code-switching of real speech.
Which regions are covered?
Najdi, Hejazi, Southern, Eastern and Northern Saudi Arabic. Tell us the regions and the mix you need.
Can I hear samples first?
Yes. Request samples with the form on this page and we will send them with the full specification.
Can it be used in a commercial product?
Yes. The data is prepared for training and evaluating commercial AI systems, under a licence agreed with you.
Related datasets
Tell us what your model needs.
Request samples and the full specification. We reply within a couple of business days.
Request samples