Music Genre Classification Using Long Short-Term Memory with Gated Recurrent Unit

Abstract

The music genre classification is an important part in data retrieval for the association of increasing music collections. It is one of the challenging tasks for classifying music with reliable accuracy. Numerous methods are utilized as handcrafted features for identifying individual patterns which are unable to define the actual characteristics of music. To overcome these issues, this paper proposed a Deep Learning (DL) based technique for classifying music genre. The Long Short-Term Memory with Gated Recurrent Unit (LSTM+GRU) is used to classifying the music genre. The GTZAN dataset is used in this paper which collects 1000 clips from 10 balanced genres music data. It is preprocessed by clip splitting procedure which is utilized to convert 30s music clip into 3s which helps to enhance the classification performance. Then, the Mel Frequency Cepstral Coefficients (MFCC) is used for extracting relevant features from input signal and LSTM+GRU is used for classifying the music genres. The accuracy, recall, precision and f1-score are used for calculating model performance. The LSTM+GRU attains accuracy 93.65%, recall 92.38%, precision 92.57% and 92.18% f1-score when compared to existing techniques like ResNet50 and Convolutional Neural Network with Bidirectional GRU (CNN+BiGRU).