Classification of Pathological and Healthy Voice Using Perceptual Wavelet Packet Decomposition and Support Vector Machine

In this study, a new approach has been presented based on perceptual wavelet packet transform and support vector machines for analysis and classification of pathological and healthy voice signals. Feature extraction and development of classification algorithm play important role in the area of automatic classification of pathological and healthy voice signals. The critical sub-bands are obtained by separating pathological and healthy voice signals with perceptual wavelet packet trans- form. The energy and entropy measures are extracted at each sub-bands used for training and testing of the classifier. In the study, the VIOCED database are used and it consists of 208 voice signals which are 58 healthy and 150 pathological. Experimental results demonstrate that the proposed features and classification algorithm provide 93.1% sensitivity, 96.5% specificity and 97.1% accuracy rates and it shows that the proposed method can be used to help medical professionals for diagnosing pathological status of a voice signal.