Language-Independent Computer Emotion Recognition

S. Mitsuyoshi and F. Ren (Japan)


Emotion Recognition,Comparison of Subjective Evalu ation,Spontaneous Emotional Speech,Consciously Emo tional Speech,Emotional Parameter


There are various problems in emotion recognition. In ad dition, with neural networks, the reproduction of judgment paths is difficult and unstable. The other problem is that in subjective comparative tests of emotional speech among humans, human emotion depends on the awareness of the person and is lacking in clarity. However, in conventional emotion recognition, there is little of the speech data neces sary for precise comparison of human emotional character istics. and the human emotion recognition rates required as criteria, the distinction of speech characteristics between the actual emotion and the emotion expressed in speech, and the matching rates of subjective evaluation among hu mans clearly still do not hold up. In order to solve these problems, this paper aims for recognition performance at a practical level, and emotion recognition analysis through a more rigorous and precise comparison with humans has been carried out. In this way, for the first time an emotion recognition logic and emotion parameters that are effec tive in practice are required. Rigorous comparative tests of emotion recognition with test subjects were carried out. In the results obtained, a matching rate of 69.75% was ob tained for continuous speech with subjective evaluation by the test subjects themselves, and a matching rate of 85.71% was obtained for comparisons with the assertions of the ac tual speaker directly after the speech was performed.

Important Links:

Go Back