Research And Implementation Of Music Emotion Recognition And Retrieval System Based On Histogram Density Model

In the fast-paced and high-pressure life of modern society,the streets and alleys are full of music.In daily life,people use music to release and relieve stress and adjust their emotions.For a long time,people's demand for music is not just leisure and entertainment,but also seeking emotional resonance.However,big data era of music resources are quite rich,the emotional recognition of music can accurately and quickly find the emotional music needed.The subjectivity of the human body is very strong,the different people on the same music emotion understanding is not the same,in order to be able to accurately identify the emotion of music,music emotion recognition research needs to consider people's subjectivity,therefore,more and more researchers try to use Valence-Arousal(VA)emotion space as the emotion space of music.In addition,the study confirmed that the acoustic signal characteristics of music signal is of great significance to the study of music emotion.In this paper,a histogram density model is used for music emotion recognition.The HDM model,the emotion space of VA is quantified into G*G lattice,and the emotion distribution of music is simulated by using two-dimensional histogram density estimation.Based on the weight of the learning histogram generated by linear combination of different audio themes,the emotion prediction of music with unknown emotions in the test set is carried out.Finally,the model is evaluated according to the predicted values of the test set and the ED,MEA,RMSE of the actual values of the test set.In this paper,SVM and KNN,Ensemble learning,AEG(Acoustic Emotion Gaussians model)and HDM model five methods are used to compare the experimental results.The model adopted in this paper has no hypothesis of parameter distribution in VA space,and the model is trained quickly and accurately to predict the emotion of music.In this paper,the performance of AMG1608 emotional annotation dataset is studied.The emotional label of 1608 30 s music segments in this dataset is completed by 665 participants.The experimental results show that the HDM model has a high accuracy in emotion recognition of music,that is,a better recognition effect.At the same time,this paper also designs and implements a music emotion retrieval system based on Android.Using MATLAB to complete the realization process of HDM model algorithm,the HDM model is trained on the basis of AMG1608 data set,and then the Android music retrieval system is developed.The Android is connected with the MATLAB,and the emotion retrieval function of the music retrieval system is realized by matching the emotional result of the music with the emotion category selected by the user.