A study of support vector machines for emotional speech recognition

In this paper, efficiency comparison of Support Vector Machines (SVM) and Binary Support Vector Machines (BSVM) techniques in utterance-based emotion recognition is studied. Acoustic features including energy, Mel-frequency cepstral coefficients (MFCC), Perceptual linear predictive (PLP), Filter bank (FBANK), pitch, their first and second derivatives are used as frame-based features. Four basic emotions including anger, happiness, neutral and sadness in Interactive Emotional Dyadic Motion Capture (IEMOCAP) database are selected for training and evaluating in our experiments. The best accuracy of emotional speech recognition is 58.40% in average from SVM with polynomial kernel. Energy features combination with FBANK, pitch and their first and second derivatives features are the most suitable for computing utterance feature. Binary Support Vector Machines (BSVM) techniques show accuracy improvement in some emotions, such as sadness and happiness emotion.