A Formula for Music Similarity: Utilising music-theoretical approaches in audible perceptions of harmonic similarity

Abstract

Harmony appears to have a vital role in listeners’ perceptions of musical similarity. However, long-established theories of harmony such as Hugo Riemann’s theory of ‘harmonic functions’ have been under-utilised in the fields of music cognition and perception, and particularly in music information retrieval and forensic musicology. Indeed, it is surprising that such crucial applications still generally rely upon ad-hoc and proprietary methods for determining similarity. My doctoral research explores whether traditional scholarly music-theoretical methods of determining harmony (such as Riemann’s theory of harmonic function, and aspects of Schenkerian analysis) could aid in developing better methods for determining similarity. I propose that we would be better able to extract high-level musical features by using traditional music-theoretical methods.

Firstly, I report an initial study that highlights harmonies relevance in participants’
classification of audible music similarity. Riemann’s theory is then utilised to explain
some of the apparent discrepancies in human-annotated harmony datasets; specifically, the Chordify Annotator Subjectivity Dataset, a subset of Chordify’s user edit data, and my own annotation study using the song ‘Little Bit O’ Soul’ (Chapters 3, 4, and 6).

This thesis concludes by proposing an adapted version of Riemannian theory (removing the need for a key), which can be applied not only to computationally encoded scores, but also audio and other computationally available data (Chapters 5, and 7). Overall, I show that a Riemannian-based approach that observes the chord labels (not using a score) enables music similarity approaches to explore audible music similarity in more depth.

This research not only has significant importance in our understanding of harmonic
similarity, but also in understanding how current audio-based extraction methods can
incorporate music theory. My use of this theoretical framework in the study of musical
similarity could improve methods of determining music similarity used in a variety of
other fields, such as the development and implication of copyright law, commercial music sales, music information retrieval extraction and evaluation metrics.

More information

Identifiers

ORCID for Nicholas Gibbins: orcid.org/0000-0002-6140-9956

Catalogue record

Export record