Music genre recognition: an explainable approach

The way music is provided and consumed has changed significantly over the past 10-15 years. Music content providers, such as Spotify, host catalogues consisting of hundreds of millions of tracks which are made available to users wirelessly and at the click of a button. As a result there are significant challenges posed in the organisation and categorisation of such catalogues. Genre, being one of the primary methods of categorising music, therefore comes to the forefront. This paper investigates the processes behind the categorisation of music into genre, known as Music Genre Recognition (MGR), using modern machine learning techniques and builds a model for this purpose. Feature extraction plays an important role with mel spectrograms and Mel Frequency Cepstral Coefficients (MFCCs) both being utilised as inputs to a Convolutional Neural Network (CNN). The novel contribution of this paper to the field of MGR is to explain the results of this model using eXplainable Artificial Intelligence techniques (XAI). A prototype MGR model was trained achieving an accuracy of 93%. The results were then interpreted using SHapley Additive exPlanations (SHAP).