The Research Of Speech Emotion Recognition Based On DBN

With the rapid development of Artificial Intelligence,Emotional Intelligence is taken more seriously as an important component of Artificial Intelligence.The goal of Emotional Intelligence is to make computer comprehend the emotion of people by analyzing the messages which come from people.Furthermore,the computer can express its emotion to people based on the analysed result.In a word,the computer itself own the emotion.Speech emotion recognition is an important component of Emotion Intelligence.Its aim is to make computer recognize the emotion type of people after receiving the speech signal.During the research process,we first extract different types of features which are related to speech emotion from speech,then we will use feature selection algorithm to realize dimension reduction for the original feature set if necessary.After getting the feature set,we use classifier to get the model.Finally,we put the test data set to the model and we will get the forecast result.Deep neural network is a hotspot of Artificial Intelligence research in the recent years.It has already got the greater achievement in the area of speech and image recognition compared to the traditional classifier.Deep neural network is a multi-layer neural network which uses big data as its input and adjust its parameter by multiple iteration.There are many familiar deep neural networks,such as Deep Belief Network,Convolutional Neural Network and so on.We choose Deep Belief Network to do the speech emotion recognition research.After getting depth understanding to the DBN and the theory of speech emotion recognition,we find that though the Deep Belief Network is suitable for the speech emotion recognition,multi-layer combination and transformation for the feature set by using Restricted Boltzmann Machine will lead to the loss of valuable features.then the accuracy will be affected inevitable.Because of this,in this paper we put forward an improved algorithm which uses random eigen subspace and multiple single-layer Deep Belief Networks as the base classifiers and uses SVM as the final classifier to get the forecast result.The result of experiment porves that the imporved algorithm gets higher accuracy compared with tranditional Deep Belief Network.