Emotion Recognition Using Voice Based on Emotion-Sensitive Frequency Ranges

Abstract

To date, study on emotion recognition has focused on detecting the values of pitch, formant, or cepstrum from the variation of speech according to changing emotions. However, the values of emotional speech features vary by not only emotions but also speakers. Because each speaker has unique frequency characteristics, it is difficult to apply the same manner to different speakers. Therefore, in the present work we considered the personal characteristics of speech. To this end, we analyzed the frequency characteristics for a user and chose the frequency ranges that are sensitive to variation of emotion. From these results, we designed a personal filter bank and extracted emotional speech features using this filter bank. This method showed about 90% recognition rate although there are differences among individuals.