Publications

Abstract : Speech is one of the most popular modalities for emotion recognition. This work uses Mel and Bark scale dependent perceptual auditory features for recognizing seven emotions from Berlin speech corpus. A combination of Mel Frequency Cepstral Coefficients (MFCC's), Perceptual Linear Predictive Cepstrum (PLPC), Mel Frequency Perceptual Linear Predictive Cepstrum (MFPLPC) and Linear predictive coefficients (LPC) are chosen for the task. The role of perceptual based features is analyzed for effective Speech Emotion Recognition (SER). A search for a compact feature vector of perceptual features is carried out in the discrete and continuous emotion category. Neural network classifier is employed for classification. Comparative analysis with literature shows that the proposed algorithm outperforms in terms of recognition accuracy. Promising results are obtained with the mentioned combination of features in discrete emotional category as well as in arousal dimension of the continuous emotion circumflex.