Emotion Recognition Of Speech And Its EEG Signals Based On Compressed Sensing

In recent years,with the continuous development of artificial intelligence technology,the requirements for human-computer interaction are getting higher and higher.As an important module of artificial intelligence technology,emotion recognition has become one of the hot research fields.In order to solve the problem of high computational complexity and noise sensitivity of recognition network in emotion recognition,an emotional recognition model based on compressed sensing(CS)was proposed in this paper.In addition,in view of the fact that a single signal can not fully represent the emotional state,and when the characteristics of a channel are interfered or missing,the result of emotion recognition will be significantly reduced,the EEG and speech fusion features were used for recognition.The main research work of this paper is as follows:The basic composition of emotional recognition system was introduced,The TYUT2.0 emotional speech database used in the experiment and the colleted emotional EEG database which stimulated by the former were introduced in detail.In addition,the preprocessing process of the two databases and the feature parameters commonly used in emotion recognition were introduced.The BP,SVM and SRC recognition models commonly used in emotion recognition were briefly explained,the advantages and disadvantages of these models were summarized,and a more robust and efficient CS emotional recognition model was proposed.In this model,the redundant dictionary of signals was constructed by K-SVD method,and the Gaussian matrix with random distribution was selected as the observation matrix.In the selection of reconstruction algorithm,an improved variable step size adaptive matching pursuit algorithm was proposed,and through the analysis of the performance of speech signal reconstruction,it was proved that the improved algorithm is superior to the traditional OMP and SAMP algorithms in reconstruction accuracy.In order to verify the anti-noise performance of CS and its effectiveness for speech and EEG emotion recognition,the emotion recognition experiments of noisy speech with different signal-to-noise ratio were carried out based on CS recognition model,at the same time,three recognition models: BP,SVM and SRC,were set up for comparison.The results showed that the CS recognition model has a more robust and superior recognition effect than other recognition models for noisy speech.At the same time,it was verified again that the CS-based emotional recognition model is superior to the other three recognition models in recognition rate and training speed through the emotional recognition experiment of pure speech signals and pure EEG signals.In order to construct a more effective emotional recognition system,thefusion features of EEG and speech were used for recognition.Based on the principle of CS reconstruction according to the interatomic correlation of samples,the canonical correlation analysis(CCA)method was used to fuse the speech and EEG features,and the serial fusion was used as a comparison.The results showed that the recognition rate based on CS for CCA fusion is 95.14%.Regardless of the average recognition rate or the single emotional recognition rate,the recognition results of CCA fusion are better than those of single speech,single EEG and serial fusion.