Vocal fold pathology detection using modified wavelet-like features and support vector machines

Acoustic analysis is a perspective vocal pathology diagnostic method that can complement (and in some cases replace) other methods, based on direct vocal fold observation. There are different approaches and algorithms for feature extraction from acoustic speech signal and for making decision on their basis. While the second stage implies a choice of a variety of machine learning methods (SVMs, neural networks, etc), the first stage plays crucial part in performance and accuracy of the classification system, providing much more creativity in development of different feature extraction methods. In this paper we present initial study of feature extraction based on wavelets and pseudo-wavelets in the task of vocal pathology diagnostic. A new type of feature vector, based on continuous wavelet and wavelet-like transform of input audio data is proposed. Support vector machine was used as a classifier for testing the feature extraction procedure. The results of our experimental study are shown.