Musical Genre Recognition using Convolutional Neural Networks
In this paper we present an overview of Musical Genre Recognition (MGR) using Convolutional Neural Networks (CNNs). We discuss the background of MGR and CNNs, before going on to discuss and compare the use of spectrograms and raw audio as the input to these models. We also discuss deconvolution of the CNNs and auralisation of the resulting spectrograms, finding that CNNs are extracting meaningful features for the task of MGR.
We conclude by finding that though CNNs are capable of extracting musical features from raw audio, they perform significantly better when extracting features from spectrograms.