Classification of voice aging using parameters extracted from the glottal signal

Abstract

Classification of voice aging has many applications in health and geriatrics. This work focuses on finding the most significant parameters to identify voice aging. This work proposes to choose the most significant parameters extracted of the glottal signal to identify the voice aging process of men and women using the wrapper approach combining a genetic algorithm (as a search algorithm) with a neural network (as an induction algorithm). The chosen parameters will be used as entries in a neural network to classify male and female Brazilian speakers in three different age groups, which will be called young (from 15 to 30 years old), adult (from 31 to 60 years old) and senior (from 61 to 90 years old). The voice database used for this work was composed by one hundred twenty Brazilian people (male and female) of different ages. In this work we use the largest basis for classification of age compared with other similar works, and its rate of classification is superior to other studies reaching 91.6% in males and 83.33% in women. © 2010 Springer-Verlag Berlin Heidelberg.