A clustering-based approach to automatic harmonic analysis: an exploratory study of harmony and form in Mozart’s piano sonatas

Abstract

We implement a novel approach to automatic harmonic analysis using a clustering method on pitch-class vectors (chroma vectors). The advantage of this method is its lack of top-down assumptions, allowing us to objectively validate the basic music theory premise of a chord lexicon consisting of triads and seventh chords, which is presumed by most research in automatic harmonic analysis. We use the discrete Fourier transform and hierarchical clustering to analyse features of the clustering solutions and illustrate associations between the features and the distribution of clusters over sections of the sonata forms. We also analyse the transition matrix, recovering elements of harmonic function theory.