Quranic reciter recognition: A machine learning approach

Abstract

Recitation and listening of the Holy Quran with Tajweed is an essential activity as a Muslim and is a part of the faith. In this article, we use a machine learning approach for the Quran Reciter recognition. We use the database of Twelve Qari who recites the last Ten Surah of Quran. The twelve Qari thus represents the 12-class problem. Two approaches are used for audio representation, firstly, the audio is analyzed in the frequency domain, and secondly, the audio is treated as images through Spectrogram. The Mel Frequency Cepstral Coefficients (MFCC) and Pitch are used as the features for model learning in the first case. In the second case of audio as images, Auto-correlograms are used to extract features. In both cases, the features are learned with the classical machine learning which includes the Naïve Bayes, J48, and the Random Forest. These classifiers are selected due to their overall good performance in the state-of-the-art. It is observed that classifiers can efficiently learn the separation between classes, when the audio is represented by the MFCC, and the Pitch features. In such a case, we get 88% recognition accuracy with the Naïve Bayes and the Random Forest showing that Qari can be effectively recognized from the recitation of the Quranic verses.