Music Genre Classification System Using Deep Learning and Phase Space Reconstruction

Abstract

In this paper, we propose a music genre classification technique using deep learning. For signal processing, we use deep learning, wavelet transform, and phase space reconstruction. First, we perform a preprocessing task to remove noise from data using wavelet transform. Second, we use phase space reconstruction to convert one-dimensional signals of music data into two-dimensional signals to generate images. Third, we inject imaged features into each training data and test data to perform deep learning training. We can predict the learning results when applying the phase space reconstruction data feature extraction process, and measure the loss and accuracy to determine the classification performance. Therefore, we can verify the performance of the proposed technique to automatically classify music data, which is increasing in the multimedia content society.