Classification of Indonesian Music Using the Convolutional Neural Network Method

Abstract

Music has a variety of genres, namely pop, rock, jazz, and so on. Indonesia has its own music that other countries do not have, including campursari, dangdut, and keroncong music. The three types of music have musical instruments that are almost similar, which makes it difficult for listeners to distinguish the genre of music, especially the younger generation, so we need a tool called classification. This study uses a mel-spectogram and the Convolutional Neural Network (CNN) method to classify Indonesian music. The CNN parameters and architecture tested in this study were batch normalization, ReLU activation, dropout, activation of sigmoid and softmax output, epoch value, learning rate value, and dense layer value. The entire parameter is tested using input with two different data sharing methods, namely stratified split and k-fold cross validation. The highest accuracy of 82% was obtained by using the stratified split data distribution method and using batch normalization parameters, ReLU activation, activation of outputs sigmoid and softmax, 30 epoch values, 0.05 learning rate values, and 200 layer dense values. The model with the highest accuracy value is used as the basis for classifying Indonesian music into campursari, dangdut, or keroncong classes.