Audio-Visual Emotion Recognition Based on Facial Expression and Affective Speech

In this paper, the performance of audio-visual emotion recognition integrating facial expression and affective speech is investigated. The local binary patterns (LBP) features are extracted for facial image representations for the single facial expression recognition. Three typical acoustic features including prosody features, voice quality features as well as the Mel-Frequency Cepstral Coefficients (MFCC) features are extracted for the single speech emotion recognition. Then, we fuse the two modalities, i.e. facial expression and affective speech, and performed audio-visual emotion recognition at the feature-level. The support vector machines (SVM) is used for all the emotion classification. Experimental results on the publicly available eNTERFACE' 05 emotional audio-visual database demonstrate that the presented method of audio-visual expression recognition obtains an accuracy of 66.51%, giving better performance than the mono-modality.