Acoustic Modeling for Indexing and Retrieval of Quranic Verses

Abstract

The Holy Quran, revered as the singular scripture of the universe and preserved in its original entirety since its divine revelation, holds profound significance within the Muslim community. Originally shown in the Arabic language, practitioners must understand and adhere to the prescribed methods of recitation and memorization as defined by native Arabic speakers. Despite the advancement of AI technology in acoustic modeling, the intricate nature of Arabic, with its diverse accents and dialects, poses a formidable challenge for developing a resilient model for Quranic recitation. Our research addresses this challenge by introducing a deep learning model that withstands linguistic variations and stays unaffected by diverse recitation styles and the nuances of the Tajweed. In this paper, the deep features extracted from this model prove exceptional performance, achieving a remarkable accuracy of approximately 96.30% in classification tasks. To underscore the significance of our deep learning network as an acoustic model, we developed a content-based verse retrieval system (CBVeRse). Utilizing the previously trained model, this system exhibited an impressive performance with a mean Average Precision (mAP) of 96.52%. This underscores the efficiency and importance of our approach in enhancing the understanding and application of the Holy Quran's acoustic attributes.