Arabic children’s read speech data for reading apps
Children reading Arabic aloud, from single letters to short stories, with the hesitations and mistakes that reading tutors must hear. Every recording is made with a parent’s consent and supervision, and every misread letter, skipped word and self-correction is marked.
Specification
- Speakers
- Children aged 5 to 12, balanced across age and gender, in Saudi Arabia, Jordan and Egypt
- Material
- Letters with harakat, words, sentences and short graded stories
- Audio
- 16 kHz, 16-bit WAV, recorded at home and in classrooms
- Annotation
- The text read, what was actually said, and each error: substitution, skip, insertion, self-correction
- Consent
- Recorded with a parent’s or guardian’s informed consent and under their supervision
- Quality check
- Reviewed by teachers of Arabic
Example of the data
ذَهَبَ الوَلَدُ إلى المَدرَسَةِ
The boy went to school (the text)
What it is used for
Reading tutors
Listen to a child read and point at the exact letter that went wrong.
Fluency assessment
Measure reading speed and accuracy the same way for every child.
Child speech recognition
Recognise young voices that adult-trained models miss.
Pronunciation games
Build games that react to how a child says each letter.
Questions
How is consent handled?
Every child is recorded with a parent’s or guardian’s informed consent, under their supervision, and no names or faces are collected.
Who marks the reading errors?
Teachers of Arabic review every recording and mark each error by type.
Can I see samples first?
Yes. Request samples with the form on this page and we will send them with the full specification.
Related datasets
Tell us what your model needs.
Request samples and the full specification. We reply within a couple of business days.
Request samples