Bridging Traditional Islamic Scholarship and Modern AI: A Human-Centered Voice Recognition System for Quran Reciter Identification

Abstract

Identifying Quran reciters from short audio excerpts can support reciter discovery, religious learning, and user engagement, yet practical deployment is challenging because recordings are often captured in noisy everyday environments. This paper presents a human-centered voice recognition system for Quran reciter identification that combines audio preprocessing, feature engineering, neural classification, and mobile deployment into a single user-facing service. The workflow begins with Quran-recitation metadata linked to online audio sources, followed by download, cleaning, duplicate filtering, normalization, fixed-length segmentation, and extraction of Mel-Frequency Cepstral Coefficients (MFCCs) and embedded utterance features. The learning task was formulated as reciter classification from short clips, with experiments comparing 10-, 15-, and 20-second durations as well as artificial neural network (ANN) and recurrent neural network (RNN) models. The results showed that a 15-second clip provided the best balance between usability and recognition capability. The ANN outperformed the RNN and was selected as the final deployed model, achieving 97.8% accuracy on the test set and 99.0% on the training set under controlled conditions. A mobile application and server-side inference service were then implemented to deliver the system to users. Although deployment remained constrained by the mismatch between clean training audio and noisy real-world recordings, the study demonstrates the feasibility of a practical, human-centered AI system for Quran reciter identification.

Keywords: Quran reciter identification, speaker identification, acoustic feature extraction, artificial neural networks

DOI: 10.54941/ahfe1007261