Automatic analysis of emotional intonation in Brazilian Portuguese

Abstract

This study analyzes emotional speech (anger, sadness, and neutral) in Brazilian Portuguese by testing 13 acoustical parameters, which were automatically processed by the software ExProsodia. The data analyzed consisted of excerpts of spontaneous speech of male and female subjects which were collected on Internet. The software selected Intonation Units (IU) based on fundamental frequency (FO), duration, and intensity information. A pilot test was carried out and its result showed that five out of 13 parameters analyzed were statically significant: skewness of fundamental frequency—sF0; coefficient of variation of fundamental frequency—cvF0; difference between medium tone and IU value—dMTIU; standard deviation of intonation unit—sdIU; and coefficient of variation of positive focus/emphasis values (above MT)—cvposF/E. A two-way ANOVA test showed a significant result. A Tukey's HSD test pointed out to differences between anger and other emotions in both genres. A significant difference between genres was found only in sadness. A Cluster test was performed based on the acoustic parameters taken into consideration and showed that only anger (male and female) could be described based on these parameters. Regarding gender differences, only sadness and neutral speech could be clustered. The results indicate that the automatic analysis done by ExProsodia can be a reliable way to analyze emotional speech produced by male and female subjects.