Research On Speech Emotion Recognition Based On Deep Learning And Fuzzy Theory

Speech is one of the most convenient communication ways among human beings.Human beings can express their emotions by speech.With the rapid development of modern science and technology,the service robots and the human beings become more and more closely together.How to make the service robots recognize the human being's emotions and respond correspondingly is the hot issue in the human-machine interaction and robot fields.Deep learning is a hot issue of the artificial intelligence fields,and it has achieved a great success since its powerful ability of feature learning.This thesis investigates the stacked autoencoders(SAE)and the convolutional neural network(CNN),and uses these two deep learning models to recognize the speech emotion.Besides,considering the fuzziness of the human being's emotions may influence the accuracy of the speech emotion recognition,the fuzzy theory is used to deal with the fuzziness.This thesis proposes a new method that combines the deep learning model and the fuzzy theory.This new method uses the structure of the fuzzy neural network,but its fuzzy membership function is replaced by the deep learning model so that the powerful feature learning ability of the deep learning model can be fully used.Besides,the fuzzy rule inference structure of the fuzzy neural network in this new method can further optimize the feature learned from the deep learning model to deal with the fuzziness of the speech emotion.In order to verify the effectiveness of the proposed method,many experiments of the stacked autoencoders and the convolutional neural network are implemented on CASIA speech emotion database to verify the results of the different structures and parameters to recognize the speech emotion.It is known from the experiment results that the proposed new method is effective,and the deficiencies are also analyzed.