Multi-features Integration for Speech Emotion Recognition

Hongjun Li, Ze Zhou, Xiaohu Sun, Chaobo Li

Abstract: Speech not only conveys the content information but also reveals the emotions of speakers. In order to achieve effective speech emotion recognition, a novel multi-features integration algorithm has been proposed. The statistical Mel frequency cepstrum coefficient (MFCC) features are directly evolved from the original speech. To further mine more useful information among statistical features, sparse groups are presented to extract the discriminative features. For enhancing the nonlinearity of features, we map features to nonlinear space to obtain nonlinear features by the orthogonal matrix. Multiple features integrated enable them to work for speech emotion recognition together. Extensive experiments comparison with state-of-the-art algorithms on CASIA dataset confirm that our algorithm can achieve effective and efficient speech emotion recognition. In addition, the analysis of different features indicates multi-features integration is superior than single type of features, where the MFCC features contribute greater in recognition accuracy and at the same time it also takes more time for features extraction.