Real-time audio-based training system for proper Qur'anic letter pronunciation

Abstract

Recitation of the Quran is an essential activity for every Muslim to understand the message from Allah to His servant. Al-Quran is written in the Arabic language, and it is important to recite it as it was written, based on what has been practiced by Prophet Muhammad s.a.w. However, this task is a great challenge, especially for those of non-Arab descent. The face-to-face traditional and prevalent method of teaching and learning the Quran with Tajweed rules starts at early ages and is time-consuming as it requires extensive practice sessions with a qualified teacher. The teacher can only see the recitation or the pronunciation of the student by looking at the face correctly and by listening, then making the corrections based on his experiences immediately. In learning the Quran, knowing the unique articulation point (Makhraj) and special characteristics (Sifaat) are the basic but essential things emphasised significantly. Although the traditional method in Quranic teaching and learning is accepted worldwide, particularly in Muslim populations, this demand for qualified teachers may not be fulfilled in many places. This issue can be overcome with an efficient learning platform to complement the existing conventional technique employing the computer and technology. Previous literature shows no similar approach highlighting the efficient Quranic learning system focusing on the basic Makhraj and Sifaat but instead concentrating on the accuracy of the Quranic verses as a whole that leads to the unsolvable problem of Tajweed. Therefore, this research embarks on developing real-time Quranic teaching and learning interactive platform, known as Computer-assisted pronunciation training (CAPT) systems, to serve as a complementary tool to systematically help Muslims recite the Quran, as a significant solution for Tajweed teaching and learning. The research was started with modelling the correct pronunciation of each letter based on the speech acoustic recorded from the experts in the Quran. Then the investigation of the combinations of the unique features of each letter concerning Makhraj and Sifaat was conducted. For Sifaat features representation, the results showed that the combinations of Mel-frequency cepstral coefficients (MFCC), and perceptual linear prediction coefficients (PLP) were the best for identifying the Sifaat of the Quranic letters. On the other hand, the combination of Mel-frequency cepstral coefficients (MFCC) and the linear prediction cepstral coefficients (LPCC) was the best way to identify the Makhraj of the Quranic letters. The process was continued with the analysis of Sifaat then Makhraj, where weighted k-nearest neighbours (KNN), medium Gaussian support vector machine (SVM), and random under-sampling boosted trees (RUSBoosted) were selected for different conditions. In this research, five classification models were developed to evaluate five pairs of the Sifaat. Another four classification models were developed to evaluate the main four Makhraj of the Quranic letters. Once the representation of the letter was completed and ready, Matlab Application Designer was used to design and build the real-time Quranic teaching and learning platform. It is very important for the proposed system to successfully work in real-time as it should mimic the process of the conventional Quranic teaching and learning where the learning happens in real-time, with immediate feedback is given to the student for improvement. To ensure the consistency of the system’s accuracy, two levels of evaluation were conducted. The first one was to test each classifier alone with a new dataset, where the classifier models have shown a good performance ranging from 70% to more than 90%. The second evaluation was conducted on the real-time system developed, where the results of the system were compared to the evaluation of human experts. The system’s accuracy score was good, where the accuracy was about 88%. The system’s ability to identify good pronunciation has also outperformed with 92%, and its ability to categorise the wrong pronunciation as "incorrect" was good, where the accuracy was about 76%. Therefore, the system developed has successfully represented the correct pronunciation of all of the Quranic letters based on Makhraj and Sifaat on an interactive computer-assisted pronunciation training, which will be a significant platform as a complementary tool for conventional Tajweed teaching and learning.