Speech emotion classification with the combination of statistic features and temporal features

For classifying speech emotion, most previous systems used either statistic features or temporal features, exclusively. However, these two distinct feature representations appear to be concerned with different aspects of emotion, and should be combined in the task. This paper proposes a classification scheme that enables the combination of them. In the scheme, GMM and MM are first performed to model the statistic features and temporal features respectively. Then the GMM likelihoods and MM likelihoods are used as features in further procedure. Finally, Weighted ayesian Classifier and MLP are applied to accomplish the classification. Experiments on Chinese speech corpus have demonstrated that the scheme could improve the classification accuracy greatly. More detailed analysis indicated that these two feature representations could compensate each other efficiently in the classification.