Multidimensional classification of normal voice qualities

Abstract

This study investigates and acoustically defines some of the perceptual parameters used to distinguish among normal male and female voices. Twenty normal speaking subjects, ten male and ten female, produced speech segments which were paired, randomized, and presented to ten listener judges. The judges rated the similarity of all combinations of voices on a seven-point scale, and these similarities were analyzed by a multidimensional analysis technique, INDSCAL. For each speaker, eight acoustical measures were obtained, and psychophysical ratings of four commonly used descriptive terms were made by ten speech pathologists. These measurements and ratings provided the interpretation of the INDSCAL solution and were also submitted to correlational analyses. The results indicate that, when distinguishing among the voices, the judges gave the greatest weight to the male versus female dimension and its acoustical correlates. The results of both the INDSCAL analysis and the correlational analyses suggest that listeners may use different perceptual strategies to classify male voices than they use to classify females ones.