Speech Emotion Feature Selection Based On Filtering And Heuristic Search

Speech emotion recognition is an important branch of human-computer interaction and intelligent research in the field of artificial intelligence.Speech emotion recognition mainly includes the establishment of speech library,the extraction of emotional features,the selection of speech features,and the recognition of emotions.The extraction of emotional features and the selection of speech emotion features are important links,which have an important impact on the level of speech emotion recognition.Selecting the characteristics of the speech emotion can effectively reduce the redundant data to select better features that characterize the speech emotion,so as to improve the recognition rate of the speech emotion.The main research work of this paper is as follows:Firstly,we establish a phonetic sentiment library,according to the TV series "The Empty Nest Grandpa" video,and obtain the phonetic emotional corpus,and build the old people's voice emotion library.Secondly,the feature selection method can be divided into filtering and encapsulation according to the different evaluation functions.In this paper,the information gain in the filtering method is selected to obtain a part of the feature set,and then the features are further filtered according to the search strategy.The search strategy includes a global optimal search class,a random search class,and a heuristic search class.Because the global optimal search time complexity is very high,it is difficult to apply in practice,and the result of random class search strategy is relatively uncertain.Therefore,the sequence forward selection algorithm in heuristic search algorithm is chosen in this paper.In this paper,the information gain and sequence forward selection algorithm are used to extract the speech emotion features on two different corpora to obtain three sets of original feature sets.And then under the information gain,sequence forward algorithm,and information gain plus sequence forward selection algorithm,we use support vector machine to acquire speech emotion recognition.The experimental results show that the feature selection method proposed in this paper can significantly reduce the dimension of the feature subset,and the recognition rate is higher than the original data.