Using deep learning for automatically determining correct application of basic quranic recitation rules

Quranic Recitation Rules (Ahkam Al-Tajweed) are the articulation rules that should be applied properly when reciting the Holy Quran. Most of the current automatic Quran recitation systems focus on the basic aspects of recitation, which are concerned with the correct pronunciation of words and neglect the other Ahkam Al-Tajweed that are related to the rhythmic and melodious way of recitation such as where to stop and how to “stretch” or “merge” certain letters. The only existing works on the latter parts are limited in terms of the rules they consider or the parts of Quran they cover. This paper comes to fill these gaps. It addresses the problem of identifying the correct usage of Ahkam Al-Tajweed in the entire Quran. Specifically, we focus on eight Ahkam Al-Tajweed faced by early learners of recitation. In the first part of our work, we used traditional audio processing techniques for feature extraction (such as Linear predictive Code (LPC), Mel-Frequency Cepstral Coefficient (MFCC), Wavelet Packet Decomposition (WPD) and Markov Model based Spectral Peak Location (HMM-SPL)) and classification (such as k-Nearest Neighbors (KNN), Support Vector Machines (SVM), and Random Forest (RF)) on an in-house dataset of thousands of audio recordings covering all occurrences of the rules under consideration in the entire Holy Quran by different reciters of both genders. In this part, we show how to improve the classification accuracy to surpass 97.7% by incorporating deep learning techniques. Specifically, this result is obtained by incorporating most traditional features with ones extracted using Convolutional Deep Belief Network (CDBN) while the classification is performed using SVM.