Finding important sound features for emotion evaluation classification

Abstract

Emotions are mental states that can be expressed by motion, speech and other physiological reactions. In human-to-human interaction emotion perception is the perception on the emotion of the other people, which, due to the nature of emotions is not so precise. On the other hand, perception on emotions in human-computer interaction is still an open problem. A lot of work is done in direction of finding suitable model for perceiving emotions based on different input signals and classification models. Here, only sound signals are considered. Still, the percentage of the classification of emotion in natural environment isn't satisfactory. Finding a suitable model for emotion classification based on emotion evaluation is the objective of this paper. We investigated the available methods for finding the most important sound features and introduced a novel approach to finding them. Our approach includes knowledge from psychological studies that analyzed the human perception on emotions. A classifier based on the features selected with the new approach is introduced and evaluated in comparison to others. The future usage and improvement on the emotion classifier build is also examined.