Explaining harmonic inter-annotator disagreement using Hugo Riemann's theory of ‘harmonic function’

ABSTRACT Harmonic transcriptions by ear rely heavily on subjective perceptions, which can lead to disagreement between annotators. The current computational metrics employed to measure annotator disagreement are useful for determining similarity on a pitch-class level, but are agnostic to the functional properties of chords. In contrast, music theories like Hugo Riemann's theory of ‘harmonic function’ acknowledge the similarity between chords currently unrecognised by computational metrics. This paper, utilises Riemann's theory to explain the harmonic annotator disagreements in the Chordify Annotator Subjectivity Dataset. This theory allows us to explain 82% of the dataset, compared to the 66% explained using pitch-class based methods alone. This new interdisiplinary application of Riemann's theory increases our understanding of harmonic disagreement and introduces a method for improving harmonic evaluation metrics that takes into account the function of a chord in relation to a tonal centre.