AI-Powered Real-Time Tajweed Mispronunciation Detection and Corrective Feedback

Learning to recite the Qur’an with proper Tajweed is challenging, as it requires precise control over pronunciation, vowel length, and pausing. Many learners struggle, especially without regular access to skilled teachers who provide immediate corrections. While traditional methods are effective, they are often constrained by time and distance and lack personalized feedback. Rapid developments in artificial intelligence (AI), particularly in speech recognition and deep learning, offer the potential to make Tajweed learning more interactive and accessible to a wider audience. However, existing approaches are limited in their coverage of Tajweed rules, often focusing on a narrow subset of errors and lacking real-time, interpretable feedback. To address these limitations, this research proposes a novel Intelligent Tutoring System (ITS) that integrates advanced acoustic modeling with explainable AI techniques. The system follows a structured pipeline where recitation audio is preprocessed and transformed into discriminative features such as Mel Frequency Cepstral Coefficients (MFCCs). These features are then processed using Transformer-based models to capture temporal dependencies in recitation, enabling accurate detection of errors related to duration-based rules (e.g., Madd and Ghunnah) as well as pausing rules (Waqf). The system emphasizes access to diverse, well-labelled datasets covering different ages, genders and Tajweed levels to avoid bias and to ensure fair performance. Beyond simply flagging errors, it provides clear explanations, helping learners improve their recitation more efficiently while serving as a complementary tool to human teachers.