Abjad-Kids: An Arabic Speech Classification Dataset for Primary Education

Abjad-Kids is an Arabic speech classification dataset designed for primary education applications. It contains spoken recordings of the Arabic alphabet, numbers, and colors from multiple child speakers, supporting research in automatic speech recognition, audio classification, and educational technology for Arabic-speaking children.

This dataset is related to the work presented in:

Abjad-Kids: An Arabic Speech Classification Dataset for Primary Education
ResearchGate. https://www.researchgate.net/publication/401732601

Dataset Description

The dataset is organized into three main categories, each with subcategories corresponding to the spoken label:

Category	Description	Example Labels
alphabet	Arabic letter pronunciations	Alam (أ), Ba (ب), Ta (ت), ...
numbers	Arabic number names	Arbaa (أربعة), Whaed (واحد), Ethenen (اثنين), ...
colors	Arabic color names	Abyad, Ahmar, Akhdar, ...

Dataset Structure

.
├── alphabet/          # Letter recordings
│   ├── Alam/
│   ├── Ba/
│   └── ...
├── numbers/           # Number recordings
│   ├── Arbaa/
│   ├── Whaed/
│   └── ...
├── colors/            # Color recordings
│   ├── red/
│   ├── blue/
│   └── ...
├── alphabet.csv       # (audio, label) metadata
├── colors.csv         # (audio, label) metadata
└── numbers.csv        # (audio, label) metadata

Data Fields

Each CSV file has two columns:

Column	Type	Description
audio	string	Path to the audio file (e.g. alphabet/Alam/001.wav)
label	string	The spoken word or letter (e.g. Alam, Arbaa, red)

Intended Use

• Automatic Speech Recognition (ASR) for Arabic children's speech
• Audio classification tasks (letter, number, color recognition)
• Educational applications for primary education and language learning
• Pronunciation assessment and computer-assisted pronunciation teaching (CAPT)

Usage

Load with Hugging Face Datasets

from datasets import load_dataset

# Load by category
alphabet_ds = load_dataset("your-username/Abjad-Kids", data_files="alphabet.csv", split="train")
numbers_ds = load_dataset("your-username/Abjad-Kids", data_files="numbers.csv", split="train")
colors_ds = load_dataset("your-username/Abjad-Kids", data_files="colors.csv", split="train")

Load with pandas

import pandas as pd

alphabet = pd.read_csv("alphabet.csv")
numbers = pd.read_csv("numbers.csv")
colors = pd.read_csv("colors.csv")

Citation

If you use this dataset in your research, please cite the related paper:

@article{abjadkids2026,
  title={Abjad-Kids: An Arabic Speech Classification Dataset for Primary Education},
  author={Snoubara, Abdul Aziz and Al-Maradni, Baraa and Al-Naal, Haya and Al-Madrmani, Malek and Jdini, Roaa and Zarzour, Seedra and Al Jallad, Khloud},
  year={2026},
  url={https://www.researchgate.net/publication/401732601_Abjad-Kids_An_Arabic_Speech_Classification_Dataset_for_Primary_Education}
}

License

MIT License