Research On Speech Emotion Recognition Based On Spectrum Perception

Speech is the most important way for humans to communicate and interact with each other,because the speech signal not only contains a variety of rich semantic information,but also conveys the rich emotional state that people communicate.The computer further analyzes the emotional characteristics contained in the speech signal and understands the emotional information contained therein for the purpose of more friendly,efficient and convenient human-computer interaction.It has important application value and great research significance.However,according to the lack of new features that are more closely related to emotional expression in existing speech emotion recognition,this article reviews a large number of relevant literature and literature abroad,and studies and further studies related theory and technology of speech emotion recognition.This paper proposes a new spectrum sensing-based subband-sensing spectral energy feature BPSE,and adopts a feature fusion algorithm to fuse the MFCC and BPSE features to obtain the new features of BPSE-MFCC to improve the performance of speech emotion recognition.The main tasks are as follows:Firstly,the commonly used feature extraction of existing speech emotions is based on the physical acoustic characteristics,only considers the physical characteristics of the sound,and there are problems such as low recognition rate.This paper proposes a subband-perceived spectral energy feature BPSE and solves problem that there is a lack of new features that are more closely related to emotional expression inemotion recognition;Secondly,For the problem that the BPSE of the sub-band perceived spectral energy of the new speech emotion feature is still relatively simple and the recognition rate is not yet optimal,the feature selection and fusion method of speech emotion recognitionfeature is adopted,and the feature parameters of speech emotion recognition using F ratio and D ratio are evaluated.In this way,we fuse the optimal features of MFCC and BPSE,and obtain a new speech emotion fusion feature BPSE-MFCC.The new feature can effectively express the physical and auditory perception characteristics of speech emotion recognition;Thirdly,a speech emotion recognition system based on the SVM model was constructed.Experiments were performed on the Chinese sentiment corpus CASIA and the Berlin German emotional speech library EMO-DB under the Matlab simulation environment.The commonly used speech emotion features,new features BPSE and new fusion features BPSE-were extracted.MFCC,and comparative analysis of these features of speech emotion recognition performance.Experimental results show that the new features BPSE and the new fusion feature BPSE-MFCC are better than commonly used speech emotion features,which greatly improves the performance of speech emotion recognition.