Indonesian Music Genre Classification on Indonesian Regional Songs Using Deep Recurrent Neural Network Method

Abstract

Indonesia has a diverse genre of music (music genre) and is spread throughout the provinces in Indonesia. Indonesian music genres include angklung, gamelan, and kulintang. The Indonesian music genre has some similarities in terms of sound such as angklung and kulintang because it is made of wood and gamelan which has various types such as Javanese gamelan and Balinese gamelan which have a similar sound so that listeners are difficult to use to stream the right music. Because of this, the Indonesian music genre is more difficult to recognize because the instruments that build are diverse and have similarities with each other. One of the tools that can be used to facilitate the introduction of the Indonesian music genre is classification. The classification carried out needs to find the appropriate parameters to determine an accurate Indonesian music genre. In this study using the mel-spectrogramand the Deep Recurrent Neural Network (DRNN)method for music classification problems. The parameters and DRNN architecture tested are dropout value, the number of Gated Recurrent Unit(GRU) hidden layer, and the output activation. In this study using 0.25; 0.5; 0.75 as a dropout value, 4, 5, and 6 as the number of GRU hidden layer, and sigmoid, softmax as the output activation. The data used are 1000 music clips with 30 seconds duration of 192kbps quality obtained from youtube as well as compact disk (CD). The highest accuracy value of 83.28% is obtained by using a dropout value of 0.25, the number of GRU hidden layers of 6, and the output activation of softmax.