Speaker’s mental state estimation method using intensities of seven emotions calculated from acoustic features
3 , 2016-06 , 人工知能学会
Previous multi-class emotion classification method has two problems; it cannot deal with emotions except pre-defined emotions, and it cannot express situation when multiple emotions are arousing at the same time. In this paper, we apply hard and soft clustering methods into emotion classification. Hard clustering method gives information to re-classify emotion classes. Soft clustering method is valid to detect data which arouse multiple emotions. Seven types of emotion voice data (anger, anticipation, disgust, fear, joy, sadness, and surprise) are re-classified into eight clusters based on intensity values of the seven emotions calculated from acoustic features. The result of hard clustering suggested that some emotion data can be classified more finely. On the other hand, some data which have the features of multiple emotions could be found by soft clustering method.