Unsupervised Music Genre Classification with a Model-Based Approach

New music genres emerge constantly resulting from the influence of existing genres and other factors. In this paper we propose a data-driven approach which is able to cluster and classify music samples according to their type/category. The clustering method uses no previous knowledge on the genre of the individual samples or on the number of genres present in the dataset. This way, music tagging is not imposed by the users' subjective knowledge about music genres, which may also be outdated. This method follows a model-based approach to group music samples into different clusters only based on their audio features, achieving a perfect clustering accuracy (100%) when tested with 4 music genres. Once the clusters are learned, the classification method can categorize new music samples according to the previously learned created groups. By using Mahalanobis distance, this method is not restricted to spherical clusters, achieving promising classification rates: 82%.