Modelling the emotional quality of speech in a telecommunication context

Abstract

This paper presents a study of the perception, the analysis, and the modelling of the emotional quality of speech. Speech emotional quality is defined as the qualities of speech samples in terms of the emotional content that describe the listeners' global impressions as elicited by their audition. For this study, twenty professional female speakers recorded a welcome prompt of a vocal server in five elocution styles. The sound corpus was submitted to psychoacoustic tests and to signal analysis. From the psychoacoustic tests, twenty subjective criteria could be extracted that characterize the perceived emotional quality. These criteria can be used to draw perceptive portraits of the speech samples. Linear models connecting the perceptive portraits to physical data derived from signal analysis were developed.