Phonetic-Based Arabic Text Articulation System for Children Using Machine Learning

Document Type

Conference Proceeding

Publication Date

2026

Abstract

Learning to read Arabic correctly provides a particular difficulty for young learners due to its rich phonetic structure, complex script, and diacritical marks that substantially alter pronunciation. However, many students struggle to evaluate their development without direct teacher evaluation. This paper presents a machine learning-based system for automatically assessing phonetic articulation of Arabic text, primarily for educational purposes with children. The system has several stages: first, the child chooses a reading and a verse to practice. Following repeated listening and memory, the child records their reading using a microphone. The recorded audio is then subjected to two major preprocessing steps: noise reduction and segmentation, in which the audio is cleaned and separated into smaller sections to increase analysis accuracy. Next, machine learning classification model is used to compare the child’s reading to the reference audio. Finally, a feedback module provides the evaluation results, indicating whether the reading is correct and highlighting mistakes. The K-Nearest Neighbors (KNN) algorithm outperformed the other classifiers, with a final accuracy of 90%, precision and recall of 66.67%, and an F1-score of 80%, making it the best-balanced model for the task. Other models, such as MLP and Random Forest, achieved 73% and 70% accuracy, respectively, whilst SVM and Decision Tree performed at 60% and 50%. These findings indicate the suggested system’s efficiency in giving intelligent and reliable feedback to help children improve their Arabic text articulation on their own.

Link to Full Text

Share

COinS