Identification of Quran Reciters through Voice Analysis and Deep LearningMazin Mohamed Ashoor Al-Kathiri1, Abdulqader Murad Abdulqader Basalama2

Abstract

In recent years, the field of Artificial Intelligence (AI) has witnessed tremendous progress, particularly in deep learning techniques that achieve remarkable results across various domains, including image processing, speech recognition, text processing, and computer vision. This research aims to leverage the capabilities of deep learning to develop an automatic identification system for Quran reciters based on their unique voices, utilizing deep convolutional neural networks (CNNs) and Log Mel Spectrograms as audio features extractor. Our study presents a series of models designed to classify Quran reciters based on their unique vocal characteristics. To enhance accessibility for users, we developed an Android application capable of running these models offline. Our model achieved an accuracy of 98% on pure sound signals for 23 reciters. However, we discovered that simply adding background noise to pure sounds, as suggested by previous studies, was inadequate for accurately representing real-world recordings. Due to the time-consuming nature of manual recordings, we recommend developing more advanced audio noise simulations that account for common signal distortions encountered in recordings from mobile devices. This could involve synthesizing typical background noises, recording artifacts, sound echo effects, and other real-world acoustic phenomena to create more realistic training data for the deep learning models.