Self-organizing map for the classification of normal and disordered femalevoices

The goal of this research was to train a self-organizing map (SOM) on vario us acoustic measures (amplitude perturbation quotient, degree of voice brea ks, rahmonic amplitude, soft phonation index, standard deviation of the fun damental frequency, and peak amplitude variation) of the sustained vowel /a / to enhance visualization of the multidimensional nonlinear regularities i nherent in the input data space. The SOM was trained using 30 spasmodic dys phonia exemplars, 30 pretreatment functional dysphonia exemplars, 30 post-t reatment functional dysphonia exemplars, and 30 normal voice exemplars. Aft er training, the classification performance of the SOM was evaluated. The r esults indicated that the SOM had better classification performance than th at of a stepwise discriminant analysis over the original data. Analysis of the weight values across the SOM, by means of stepwise discriminant analysi s, revealed the relative importance of the acoustic measures in classificat ion of the various groups. The SOM provided both an easy way to visualize m ultidimensional data, and enhanced statistical predictability at distinguis hing between the various groups (over that conducted on the original data s et). We regard the results of this study as a promising initial step into t he use of SOMs with multiple acoustic measures to assess phonatory function .