Verbal and non-verbal cues in the communication of emotions

Several studies have shown that humans extend their interpersonal behavioral patterns onto their interaction with computers. Based on this finding, research in human-computer interaction acknowledges the need to detect the users´ expressions of emotion. However, we discovered that most of the current research is confined to emotion synthesis. In this investigation, we explored the role of verbal and non-verbal information in the communication of emotions. By training emotion-specific language and prosodic models on a corpus consisting of several thousands of sad, angry, or neutral speech segments from English movies, we showed that a classification system based on these models achieved an accuracy comparable to the accuracy of human listeners performing the same task