Accurate emotion recognition from speech is important for applications like smart health\ncare, smart entertainment, and other smart services. High accuracy emotion recognition from Chinese\nspeech is challenging due to the complexities of the Chinese language. In this paper, we explore how\nto improve the accuracy of speech emotion recognition, including speech signal feature extraction\nand emotion classification methods. Five types of features are extracted from a speech sample:\nmel frequency cepstrum coefficient (MFCC), pitch, formant, short-term zero-crossing rate and\nshort-term energy. By comparing statistical features with deep features extracted by a Deep Belief\nNetwork (DBN), we attempt to find the best features to identify the emotion status for speech.\nWe propose a novel classification method that combines DBN and SVM (support vector machine)\ninstead of using only one of them. In addition, a conjugate gradient method is applied to train DBN\nin order to speed up the training process. Gender-dependent experiments are conducted using an\nemotional speech database created by the Chinese Academy of Sciences. The results show that DBN\nfeatures can reflect emotion status better than artificial features, and our new classification approach\nachieves an accuracy of 95.8%, which is higher than using either DBN or SVM separately. Results also\nshow that DBN can work very well for small training databases if it is properly designed.
Loading....