Quantification of dysphony with allowance for inter-utterance variation

This article describes a method for the classification and quantification of dysphony caused by laryngeal pathologies. In the method described here, each case is represented by a 3-parameter feature vector in the extraction of which variation among the utterances of each given subject is taken into account by averaging over the values of the parameters obtained from three recorded utterances of the test vowel, the Spanish /a/. Quantification of dysphony is achieved by using the simple Chebychev norm to measure distance from a template representing normality, and the Mahalanobis distance is employed to distinguish 4 classes, the non-pathological and the slightly, moderately or severely pathological. The results show that the procedure is succesful in classifying, quantifying and monitoring a considerable variety of pathologies.