Speech Emotion Recognition Based On Extended Deep Belief Network

Speech signal is an important communication medium between people.Speech signal includes not only the text content that the speaker wants to convey,but also the emotional of the speaker.Artificial intelligence research is developing rapidly,and emotional speech recognition is an important research content,which has great practical significance.Emotion is a psychological process in which human brain stimulates and feels the value characteristics of things.Speech emotion recognition is a good way to make machines understand humans better.In order to identify various human emotional changes and categories,the process of recognition is mainly based on the analysis of speech signals with human emotional characteristics.The process of recognition is to use the speech signal that carry the emotional information,and through the analysis of the computer to complete the real emotions.Now,the traditional machine learning models play an important role in speech emotion recognition.However,the recognition accuracy of traditional recognition algorithms is not good enough in practical applications,and the recognition rate of algorithms still needs to be improved.With the development of deep learning technology,this paper proposes a method of speech emotion recognition based on extended deep belief networks.The specific research work is as follows:(1)Speech signal analysis and preprocessing,including the sound pressure,sound intensity,and loudness of the speech signal.Commonly used voice signal pre-processing methods,such as the pre-emphasis and de-emphasis process of the voice signal,the frame and window operation and the endpoint detection processing of the voice signal.(2)Voice signal feature extraction.Combining with the feature parameters of speech signals that are closely related to speech emotions,this paper proposes a method of using the original parameter combination to derive parameters for feature extraction of the sound signal.Including the energy parameters,pitch frequency,formant coefficient,Mel frequency cepstrum coefficient and their derived parameters.(3)Comparative experiment design based on BP neural network.Through comparative experiments using BP neural network,analyzed the recognition effect of speech emotion recognition based on BP neural network.Compared various model parameters and the model performance was continuously optimized,so that the model reached a relatively high recognition accuracy.(4)The experimental design of speech emotion recognition based on the extended deep belief network model.Using python programming language and Tensorflow deep learning framework to establish an extended deep belief network model.Through a variety of parameters,verifies the feasibility of the extended deep belief network model for speech emotion recognition.The results verify that the recognition rate of EDBN model is much higher than that of BP model and traditional machine learning model.