Dilated-Convolutional Recurent Neural Network for Music Genre Classification

Abstract

In the digital era, utilizing technology to automatically classify music genres has become very important, especially for applications such as music recommendation, music trend analysis, and digital music library management. This research evaluates the use of Dilated-Convolutional Recurrent Neural Network (D-CRNN) in classifying music genres. This method combines the advantages of Dilated-CNN in capturing longer temporal context with the temporal sequence recognition capability of CRNN. The data used is the GTZAN dataset consisting of 1,000 30-second audio recordings, categorized into 10 music genres. Data preprocessing involved converting the audio recordings into Mel-Frequency Cepstral Coefficients (MFCC) images. The model was tested using data without augmentation and with augmentation, resulting in a total of 15,991 images for training. The results show that the use of D-CRNN can improve the accuracy of music genre classification compared to the conventional CRNN method.