FCM-based multiple random forest for speech emotion recognition

Speech emotion recognition (SER) is of great importance in the human-robot interaction, and SER have been developed rapidly in recent years. How to identify high correlation features is a main question. In this paper, multiple random forest (MRF) is proposed, which can effectively predict up to several thousand explanatory variables, and improve the recognition result. At the mean time, personalized features and non-personalized features are fused for SER that can be divided into different subclasses by adopting the fuzzy C-means (FCM) clustering algorithm, then FCM-based MRF is proposed. Finally, our model is established, and the confusion matrix is output. Moreover, we conduct a separate classification of some certain emotion, which are difficult to recognition to some extent. And results show that the proposed method of FCM-based MRF achieves good performance for SER, which the recognition accuracy is 84.17%.