Unsupervised Music Genre Classification with a Model-Based Approach
Oct 2011
New music genres emerge constantly resulting from the in-
fluence 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 previ-
ous 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’ sub jective knowledge about music genres, which may also
be outdated. This method follows a model-based approach to group mu-
sic 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 cat-
egorize 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%.
