Emotion recognition for disgust and boredom states

Abstract

In this paper, we made the emotion recognition for Romanian language using EMO-IIT database with seven emotions (joy, sadness, fury, neutral tone, anxiety, disgust and boredom). Compared to our previous studies we introduced two new emotions: disgust and boredom and a new set of sentences in order to express better the emotional states. The best recognition rate of emotions is around 75% and was obtained for feature vectors which includes MFCC (Mel Frequency Cepstral Coefficients) + PARCOR (Partial Correlations Coefficients) + LAR (Log Area Ratios Coefficients). The accuracy rates are closed to the other studies from the literatures. For example, for the German emotional database EMO-DB which contains all seven emotions, the accuracy recognition rate reported by the researchers was around 85%. The disgust is often recognized as boredom (15%) or neutral tone (10%). The sadness is confused with the neutral tone (12%) and disgust (9%). The main difference between the two databases is that the EMO-IIT contains unprofessional voices with recordings provided by students and EMO-DB contains professional voices, recorded from actors.