Tag Archive Computational Analysis


Trajectory Descriptors: Music Genre Classification through the Tonnetz


We present an approach to geometrically represent and analyze the harmonic content of musical compositions based on a formalization of chord sequences as spatial trajectories. This allows us in particular to introduce a toolbox of novel descriptors for automatic music genre classification. Our analysis method first of all implies the definition of harmonic trajectories as curves in a type of geometric pitch class spaces called Tonnetz. We define such curves by representing successive chords appearing in chord progressions as points in the Tonnetz and by connecting consecutive points by geodesic segments. Following a recently established hypothesis that assumes the existence of a narrow link between the musical genre of a work and specific geometric properties of its spatial representation, we introduce a toolbox of descriptors relating to various geometric aspects of the harmonic trajectories. We then assess the appropriateness of these descriptors as a classification tool that we test on compositions belonging to different musical genres. In a further step, we define a representation of transitions between two consecutive chords appearing in a harmonic progression by vectors in the Tonnetz. This allows us to introduce an additional classification method based on this vectorial representation of chord transitions.

Video Presentation:

Conference Article:



This work has been developed as part of the doctoral studies of Christophe Weis and is published in the Proceedings of the Sound and Music Computing Conference 2024 in Porto, Portugal.