Voice-Based Hijaiyah Letter Recognition for Pronunciation Assessment Using Dynamic Time Warping (DTW)

Accurate pronunciation of Hijaiyah letters is essential in learning to read the Quran because each letter has specific phonetic characteristics and articulation points. However, conventional pronunciation assessment commonly relies on direct teacher evaluation, which may be constrained by time, assessor subjectivity, and limited access to qualified instructors. This study proposes a lightweight voice-based Hijaiyah letter recognition system for pronunciation assessment using Dynamic Time Warping (DTW). The dataset consisted of 350 audio samples, including 325 publicly available recordings and 25 recordings collected from five participants, covering five selected Hijaiyah letters: a, ba, ta, la, and ma. Audio recordings were standardized, amplitude-normalized, and trimmed to remove irrelevant silent segments. The resulting normalized one-dimensional time-domain amplitude sequences were used directly as the acoustic representation for DTW matching against class-specific reference patterns. DTW was used to compensate for temporal differences in pronunciation duration and speaking rate by finding the minimum-cost alignment between input and reference sequences. The system correctly recognized 318 of 350 evaluated samples, corresponding to an accuracy of 90.86%. The results demonstrate the feasibility of direct waveform-based DTW as a training-free and computationally simple baseline for limited-class Hijaiyah pronunciation recognition. However, its sensitivity to noise, amplitude variation, and spectral differences indicates that feature-based representations such as MFCC should be investigated to improve robustness.