Emotion recognition of mandarin speech for different speech corpora based on nonlinear features

In this paper, three speech corpora were established, three nonlinear features based on Teager Energy Operator and two linear features for emotion recognition were researched. HMM-based emotion recognition was used to evaluate the emotional recognition performance of features based on Teager energy operator. The results show that performance of two features, i.e. NFD_Mel (Nonlinear Frequency Domain based Mel-scale coefficients), AF_Mel (Amplitude-Frequency property of TEO based Mel-scale coefficients), are optimal in all the researched features. It could be therefore said that transformation of Teager Energy Operator in frequency domain, and application of amplitude-frequency property of Teager Energy Operator provide good representations of emotion styles in three speech corpora for emotion recognition.