Dimensionality reduction for voice disorders identification system based on Mel Frequency Cepstral Coefficients and Support Vector Machine

Nowadays, due to the severe daily activities and vocal abuse, many diseases affect the mechanism of voice production which causes pathological voices. Therefore, the identification of voice diseases becomes a real challenge. In this context, the automatic speech recognition can provide great results as a complementary tool to other medical techniques. This paper proposes a reliable algorithm based on short-term cepstral parameters, Linear Discriminant Analysis (LDA) as dimensionality reduction method and Support Vector Machine (SVM) as classifier. A full comparative study is established and the system performance is evaluated in terms of accuracy, sensitivity, specificity, precision and Area Under Curve (AUC). Our findings demonstrate that the detection of voice disorders can be efficient using only the original Mel Frequency Cepstral Coefficients (MFCC) ignoring their first and second derivative.