Comparative analysis between wavelets for the identification of pathological voices

Abstract

This study presents a comparative analysis of wavelets, in order to find a descriptor that provides a better classification of voice pathologies. Different types of Wavelet Packet Transform were used as a tool for feature extraction and Support Vector Machine (SVM) to classify vocal disorders. Tests were conducted with 23 wavelets types in two SVMs, the first using the strategy "one vs. all" to classify normal and pathological voices and the second, using the strategy "one vs. one" to classify pathologies: edema and nodules. The best results were obtained using Daubechies family, especially Daubechies 5 (db5) wavelet. © 2010 Springer-Verlag.