Time-frequency and phase derived features for emotion classification

Emotion recognition and synthesis plays a crucial role in Human-computer interface. In this paper, we propose a multi style emotion recognition algorithm using time frequency (pH) and phase delay of a speech signal. Most of the work done so far on emotion recognition using spectral features mainly focuses on magnitude of the signal. Phase delay has been incorporated in this work yielding better results in detecting low arousal emotions. Here, we include phase components along with the time frequency feature to form the feature vector thus increasing the efficiency by about 12%. Berlin database has been used for training and testing yielding recognition of 80.95% for seven emotions. SVM classifier is used in this work.