Impact of gender and emotion type in dialogue emotion recognition

This paper presents a dialogue emotion recognition system using Hidden Markov Model (HMM). We have compared accuracy of Mel-frequency cepstral coefficients (MFCC), Energy, and wavelet sub-band energies and their first derivative and all possible combination. Based on our experiment, MFCC show better performance in comparison with the other studied features. We have also evaluated the impact of gender and emotion states on emotion detection. Experimental results show that a significant difference is observed depending on the type of emotion studied but also in the gender of the evaluated person.