Pathological voice classifcation using local discriminant basis and genetic algorithm

Abstract

Voice feature extraction plays an important role in analyzing and characterizing voice content. The key to extract strong features that characterize the complex nature of voice signals is to identify their discriminatory subspaces. Diagnosis of pathological voice is one of the most important issues in biomedical applications of speech technology. There are some approaches for separating pathological from normal voice signals but a few ones are sophisticated to separate two or more kinds of speech pathologies from each other. This paper introduces an algorithm to discriminate voice pathologies signals from each other via adaptive growth of Wavelet Packet tree, based on the criterion of local discriminant bases (LDB). Two dissimilarity measures were used in the process of selecting the LDB nodes and extracting features from them. Moreover, Genetic Algorithm is employed for selecting the best feature set and Support Vector Machines as classifier to obtain as much as possible better results. To evaluate the proposed approach, we apply our algorithm to separate paralysis, paresis, and nodule from each other. Experimental results show the superior performance of this combinational approach against its incomplete versions (e.g. in the case of separating paresis and nodule, the proposed approach leads to 100% performance against 94.44% for where only complete wavelet packet features without applying LDB algorithm are used).