Speech Emotion Recognition Based on Deep Learning and Fuzzy Optimization

Abstract

The automatic speech emotion recognition (SER) is a challenging and attractive task. To gain a high accuracy in SER, it is need to extract salient features from the speech. Therefore, in this paper, we first verify whether the deep convolutional neural network (DCNN) can improve the salient of hand-crafted acoustic features, and then based on the conclusion of the first step, considering the fuzziness of emotion, we propose a theoretical approach of fuzzy neural network (FNN) combined with deep learning to deal with the fuzziness of emotion. The first step work is tested on the CASIA Chinese Emotional Corpus, and the second part is introduced in theory and analyzed about its feasibility.