Research On Emotional Classification Of Intangible Cultural Heritage Music Based On DBN And SVM

The intangible cultural heritage music has gradually attracted extensive attention due to its potential cultural,artistic and commercial value.With the development of national economy,art forms become more and more diverse,which brings cultural shock to the intangible cultural heritage music.Some intangible cultural heritage music even falls into the predicament of being lost.At present,there are few studies on the intangible cultural heritage music.In this paper,the high level features used for emotion classification are extracted based on the deep belief network(DBN)with the fusion features as the network input,which improves the classification accuracy of intangible cultural heritage music emotion.The main research work of this paper is as follows:(1)According to the existing music emotion model and the special background of the intangible cultural heritage music,this paper established an emotion model that conforms to the expression of the intangible cultural heritage music's emotion.According to the national intangible cultural heritage list,in this paper,1903 music fragments of the intangible cultural heritage music were collected and sorted,and a music library containing 4 emotional attributes was established.(2)According to the emotional expression characteristics of the intangible cultural heritage music,the fusion features which can reflect the tone,tone quality,timbre and time-frequency domain properties of music were extracted.In the classification experiment of music,the validity and superiority of the fusion features were proved by comparing with the traditional features extracted.(3)The fusion features are taken as input,the high-level features of music samples are extracted by using the improved deep belief network,and the emotion classification of the intangible cultural heritage music is realized by combining with support vector machine(SVM).The improvement of the deep belief network is mainly realized by adding the Dropout layer,changing the updating principle of weight and bias in the network.By comparing with the traditional classification methods,it is found that using the fusion features extracted in this paper and combining the classification model of DBN and SVM can get a better music emotion classification effect,and the classification accuracy is up to 74.3%.