Cancer larynx detection using glottal flow parameters and statistical tools

Abstract

One of the most interesting application of speech processing is the identification and classification of pathological voice. Such area of research is still a challenging task. In this paper, we are interested in the severe case of the pathological voice; larynx cancer. This work concentrates on developing a feature extraction for detecting and classifying larynx cancer by investigating different glottal flow parameters. Once the glottal flow is extracted, temporal and frequency parameters are calculated. From the large obtained set of parameters, we choose the most pertinent in terms of pathologic/normal discrimination. For this purpose, a deep analysis of statistical tools such as boxplot and probability density permits to select and ordered the most significant parameters. The detection and the classification of the larynx cancer is achieved by artificial neural network (ANN). A rate about 96.9 % of the discrimination accuracy is achieved using the developed technique.