Hybrid CNN and RNN Variant Model for Indian Music Genre Classification

Music genre classification contributes largely to information retrieval for organizing growing collections of music. The process of classification of music is very difficult and unreliable. Many methods utilized handcrafted features to identify unique patterns but were still unable to determine original music characteristics. In comparison, music classification by applying deep learning models exhibits a dynamic and effective potential. Although there are a large number of neural networks, the combination of CNN and variants of RNN has not received adequate consideration. Moreover, on the shortcomings of the specific neural network classification model, we would like to propose a hybrid architecture of CNN and variant of RNN, such as a GRU. In addition, we compare the performances based on Mel-spectrogram and MFCC features.