Laryngeal pathology detection using EMD-based voice acoustic features analysis and SVM-RBF

Abstract

Traditional techniques for detecting laryngeal pathologies, such as laryngoscopy and endoscopy, are costly and invasive. This study presents a novel approach for detecting laryngeal disorders using empirical mode decomposition (EMD)-based acoustic features analysis and support vector machine (SVM) with a radial basis function (RBF) kernel. The experiments were conducted using the Saarbrucken voice database (SVD). The voice signals were then decomposed using EMD to extract the intrinsic mode functions (IMFs). The IMF with the highest energy value was selected as the most relevant. A set of acoustic features, including mel-frequency cepstral coefficients (MFCCs), linear predictive cepstral coefficients (LPCCs), Pitch (fundamental frequency), higher-order statistics (HOSs), zero-crossing rate (ZCR), spectral centroid (SC), and spectral roll-off (SRO), is derived from the most relevant IMFs and fed into an SVM classifier to differentiate between healthy and pathological voices. Experimental results demonstrate the effectiveness of the proposed methodology, achieving a high classification accuracy of 94.5%, a sensitivity of 94.2%, a specificity of 95.3%, and an F1 score of 96.1%, outperforming conventional approaches. These results highlight the potential of EMD-based voice analysis as a non-invasive and reliable tool for early diagnosis of laryngeal disorders.