MFCC, PLP, and AI Approaches for Laryngeal Lesion Diagnosis

There is a growing global interest in the use of machine learning (ML) techniques to diagnose voice disorders. However, existing studies suffer from several limitations, such as the lack of standardized databases and the absence of frameworks for long-term patient monitoring. To address these gaps, this work introduces a national initiative to establish the first Moroccan database dedicated to throat cancer and pathological voices. The dataset includes demographic, clinical, and acoustic information collected using a unified recording protocol. The voice signals are analyzed using Mel-Frequency Cepstral Coefficients (MFCC), which are widely recognized for their effectiveness in detecting vocal abnormalities. Acoustic analysis is performed using tools developed at the Laboratory of Speech and Language (LPL) to ensure reliability and reproducibility. This initiative aims to support epidemiological research and provide clinicians with ML-based tools adapted to the local context, ultimately improving diagnostic accuracy and early detection of vocal disorders.