| People are prone to feel pressure in fast-paced work and life,resulting in anxiety.Long-term anxiety seriously affects individual social life and communication.Music is a non-invasive and effective way to affect emotions.At present,music therapy is a widely used means of decompression for anxious people.Traditional music therapy lacks the pertinence of individual’s perception of music,and manual arrangement requires a lot of professional music knowledge and is time-consuming and labor-intensive.Combining the physiological signal acquisition system and the deep learning model,this paper proposes a technology that uses multiple physiological signals to judge the individual’s perception of the music segment and feeds it back,and the computer recomposes the music to adjust the decompression technology.In order to determine the individual’s perception of music fragments,an emotion recognition model based on multi-channel physiological signals is built,with the purpose of obtaining the corresponding music fragments when the anxiety changes to calm emotions.Select EEG,ECG,and RSP that are closely related to emotions,and use the complementarity of multiple signal to construct a feature-level fusion emotion recognition model,which can effectively identify anxiety and calm emotions.The multi-channel physiological signal emotion recognition model has undergone experimental statistics.The results show that the average accuracy rate of recognizing anxiety is 82.36%,and the average accuracy rate of recognizing flat emotions is 85.73%.Compared with the single EEG signal,the accuracy of recognizing anxiety and ease emotions is increased by 6.48%,8.57%.Aiming at the existing problems of computer composition,a decompression music reconstruction model based on deep learning is built.A stacked auto encoder(SAE)is introduced into the long short-term memory(LSTM)network,and the SAE-LSTM network is proposed to reconstruct decompression music.After reconstruction,high-quality music with effective decompression can be obtained.First,calculate the spectral centroid of the corresponding music segment when the emotion is calm,and select the music segment with similar spectral centroid as the training set for reconstructing the music network.Then the training set is input into the SAE-LSTM network through One-hot encoding,and the decompression music sequence is reconstructed.Finally,the experimental results show that the SAE-LSTM network calculated by the statistical method has a higher degree of music harmony than the LSTM network,and the mean square error of the note distribution is small.In addition,the results show that SAE extracts the potential features of music,which enables the SAE-LSTM network to better learn the dependence relationship between notes.Based on the above research,the design decompression experiment is divided into a control group and an experimental group.Through the Anxiety Self-Rating Scale(SAS)and Hamilton Anxiety Scale(HAMA),the experimental evaluation of anxious people is carried out.The statistical results verify the effectiveness of the stress-reducing music reconstruction system based on multiple physiological signals. |
