NEURAL NETWORKS-BASED AUTOMATIC AUDIO CLASSIFICATION FOR AL-QURAN CHAPTERS

Abstract

Al-Quran Audio classification is one example of content-based analysis of audio signals._x000D_ This study aims to design a neural network that is able to classify Al-Quran audio files to_x000D_ the correct chapter ( سورة ), and this requires implementing state of the art Convolutional_x000D_ Neural Network (CNN) to train Al-Quran Dataset and predict the correct chapter (سورة )._x000D_ In order to achieve this aim, a critical evaluation of the current state of the automatic_x000D_ based reciting classification of Al-Quran was conducted, and the principles, assumptions_x000D_ and methods at the field were used to present a prototype based on this evaluation._x000D_ Special focus is placed upon creating a suitable robust Quranic dataset and on_x000D_ discovering the features of that dataset that make it possible for an automated recognition_x000D_ of Al-Quran chapters and recitation. In addition, it sets out principles that should be kept_x000D_ in mind when designing Al-Quran reciting recognition and learning systems, and a_x000D_ prototype based on these features is presented. The thesis provides a framework for the_x000D_ auditory classification of Al-Quran chapters, as the final results shows that the use of a_x000D_ newly created IQRA-15 dataset and CNN as a model architecture produced in excess of_x000D_ 90% accuracy on unseen data. This is a proof of concept that deep learning can achieve_x000D_ good results when applied to Al-Quran. This knowledge can be used to design an AI based_x000D_ system for self-correcting Al-Quran recitation for Arabs and non-native Arabic_x000D_ speakers.

Date of Award	2018
Original language	American English
Awarding Institution	HBKU College of Science and Engineering