An Improved Speech Emotion Recognition Algorithm Based on Deep Belief Network

Abstract

In order to improve the characterization ability of speech signal and recognition accuracy of speech emotion recognition, a speech emotion recognition model based on improved Deep Belief Network (DBN) is proposed. The method is to replace the traditional DBN activation function with a Rectified Linear Unit(Relu). And the reconstruction error is used to determine the depth of the DBN network. The short-time energy, short-time zero crossing rate, the fundamental frequency, formants and 24 dimensional MFCC parameters of emotional speech signal are extracted as the basic features. Using these basic features as input to the DBN, automatic recognition of the 6 emotions, anger, fear, joy, calmness, sadness and surprise can be achieved. Compared with the traditional DBN model and the BP model, a better recognition result is achieved by using the improved DBN discussed in this paper, and the recognition rate can reach 84.94%.