Music Genre Classification using Spectrograms
Madhur Nirmal, B. S. Shajee Mohan
Abstract
Music genre recognition (MGR) is an area of research in the broad scope of music information retrieval (MIR) and audio signal processing. Music genres are categorical labels created by humans to determine the style of music. The paper proposes a method to classify music using spectrograms. A spectrogram is the Short Time Fourier Transform (STFT) of an audio signal. They are images showing time and frequency components of an audio signal. In this work, the music signals are first converted to their corresponding spectrograms. These spectrograms are then given as input to the classifier. The classifier used in this work is a Convolutional Neural Network (CNN). Two CNN models are discussed in this paper: A user-defined CNN model and a pre-trained convnet. Pre-trained convnet makes use of the concepts of fine tuning and transfer learning. The performance of the classifier is evaluated using performance measures such as confusion matrix and classification accuracy. Three music genres such as blues, classical and rock from the GTZAN dataset are selected for experimentation. The classification accuracy is found to be good.