Characteristics with opposite of quranic letters mispronunciation detection: a classifier-based approach

Abstract

Reading Quran for non-Arab is a challenge due to different mother tongues. learning Quran face-to-face is considered time-consuming. The correct pronunciation of Makhraj and Sifaat are the two things that are considered difficult. In this paper, Sifaat evaluation system was developed, focusing on Sifaat with opposites for teaching the pronunciation of the Quranic letters. A classifier-based approach has been designed for evaluating the Sifaat with opposites, using machine learning technique; the k-nearest neighbour (KNN), the ensemble random undersampling (RUSBoosted), and the support vector machine (SVM). Five separated classifiers were designed to classify the Quranic letters according to group of Sifaat with opposites, where letters that are classified to the wrong groups are considered mispronounced. The paper started with identifying the acoustic features to represent each group of Sifaat. Then, the classification method was identified to be used with each group of Sifaat, where best models were selected relying on various metrics; accuracy, recall, precision, and F-score. Cross-validation scheme was then used to protect against overfitting and estimate an unbiased generalization performance. Various acoustic features and classification models were investigated, however, only the outperformed models are reported in this paper. The results showed a good performance for the five classification models.