There is no research for analyzing the relationship between subjectively annotated emotions and objectively annotated emotions despite the fact that a lot of research uses objective emotion labels as subjective emotions. In this study, we collect natural emotional voices, and the speaker and others annotate the intensity of six emotions {anger, dislike, fear, happiness, sadness, and surprise} to the voices. The correlation diagrams between subjectively and objectively annotated emotions indicate that there is a small relationship between subjective and objective emotions. The experimental results using support vector regression reveal that learning subjective emotions is much more difficult than learning objective emotions. Furthermore, happiness, sadness, and surprise are comparatively learned better than dislike, anger, and fear.

International MultiConference of Engineers and Computer Scientists 2019(IMECS 2019), Mar. 13-15, 2019, Hong Kong