Is it possible to represent the horizontal motions of the melodic strands of a contrapuntal composition, or the main ideas of a jazz standard as mathematical entities? In this work, we suggest a collection of novel models for the representation of music that are endowed with two main features. First, they originate from a topological and geometrical inspiration; second, their low dimensionality allows to build simple and informative visualisations.

Here, we tackle the problem of music representation following three non-orthogonal directions. We suggest a formalisation of the concept of voice leading (the assignment of an instrument to each voice in a sequence of chords) suggesting a horizontal viewpoint on music, constituted by the simultaneous motions of superposed melodies. This formalisation naturally leads to the interpretation of counterpoint as a multivariate time series of partial permutation matrices, whose observations are characterised by a degree of complexity. After providing both a static and a dynamic representation of counterpoint, voice leadings are reinterpreted as a special class of partial singular braids (paths in the Euclidean space), and their main features are visualised as geometric configurations of collections of 3-dimensional strands.

Thereafter, we neglect this time-related information, in order to reduce the problem to the study of vertical musical entities. The model we propose is derived from a topological interpretation of the Tonnetz (a graph commonly used in computational musicology) and the deformation of its vertices induced by a harmonic and a consonance-oriented function, respectively. The 3-dimensional shapes derived from these deformations are classified using the formalism of persistent homology. This powerful topological technique allows to compute a fingerprint of a shape, that reflects its persistent geometrical and topological properties. Furthermore, it is possible to compute a distance between these fingerprints and hence study their hierarchical organisation. This particular feature allows us to tackle the problem of automatic classification of music in an innovative way. Thus, this novel representation of music is evaluated on a collection of heterogenous musical datasets.

Finally, a combination of the two aforementioned approaches is proposed. A model at the crossroad between the signal and symbolic analysis of music uses multiple sequences alignment to provide an encompassing, novel viewpoint on the musical inspiration transfer among compositions belonging to different artists, genres and time. To conclude, we shall represent music as a time series of topological fingerprints, whose metric nature allows to compare pairs of time-varying shapes in both topological and in musical terms. In particular the dissimilarity scores computed by aligning such sequences shall be applied both to the analysis and classification of music.