Research On Music Genre Classification Method Based On Deep Learning

With the rapid development of the Internet,digital music has exploded.In the face of massive music,how to quickly and accurately retrieve the music that users want is becoming more and more important.As an important part of music information retrieval,music classification has become a research hotspot in recent years.The traditional music classification method mainly extracts features manually and then uses machine learning to classify.This method has two defects.One is that it is difficult to ensure the validity and accuracy of the features by manually extracting features.The second is the traditional machine learning classification.The method performs poorly on multi-classification problems and cannot train large-scale data.Aiming at the above problems,this paper proposes a method based on deep neural network for music genre classification,and studies the influence of different structures of deep neural networks on music genre classification.The main work of this paper is as follows:(1)According to the timing characteristics of music signals,a music genre classification method using long-short term memory network LSTM as a classifier is proposed.In the experiment,firstly,the characteristics of music content,such as Mel cepstrum coefficient,spectral contrast and spectral centroid,are manually extracted,and then five sets of contrast experiments are performed to find the better feature combination.The experimental results show that when the three characteristics of Mel Cepstral Coefficient,Spectrum Contrast and Spectrum Centroid are combined,the accuracy of long and short memory networks in music genre classification is higher than other feature combinations.(2)According to the advantages of convolutional neural network in image processing,a music genre classification method using sound spectrum as input data and convolutional neural network as classifier is proposed.Since the spectrogram is image data,including the frequency distribution of the music and the variation of the sound amplitude,it is very suitable for using a convolutional neural network.The convolutional neural network can automatically learn the ability of abstract features,and combine the local features of music to form a global expression.In the experiment,the convolutional neural network designed by this paper and the improved network based on VGG16 are trained firstly using the music spectral data of the music.Then,some music features are manually extracted as training data,input to support vector machine,random forest,decision tree,logistic regression and other classification models.Finally,the method based on convolutional neural network is compared with the traditional machine learning classification method.The experimental results show that the method based on convolutional neural network is superior to the traditional machine learning classification method in classification accuracy.