Feature extraction based on mel-scaled wavelet packet transform for the diagnosis of voice disorders

Abstract

Feature extraction from the vocal signal plays very important role in the area of automatic detection of voice disorders. Many feature extraction algorithms have been developed in the last three decades based on acoustic analysis of speech signals. To provide a robust representation of speech signal in an automatic voice disorder diagnosis system, a Melscaled wavelet packet transform has been suggested. A simple neural network model is designed to test the efficacy of Melscaled wavelet packet transform based features. Experimental results show that the features derived by using Melscaled wavelet packet transform provides comparable results. The proposed features can be used as an additional acoustic indicator for the evaluation of voice disorders.