Discovering Common Practice: Using Graph Theory to Compare Harmonic Sequences in Musical Audio Collections

In recent decades, rapid technological advances have resulted in a huge quantity of readily accessible digital musical recordings. The scope of large corpora presents difficulties for curators but offers new opportunities to musicologists and music theorists. We propose an application of graph theory which enables comparison of harmonic content from musical audio across collections of recordings. We introduce a graph schema wherein the chord sequences of musical recordings are used to create directed, weighted graphs which represent the underlying harmonic structure of the source material. We believe this application of graph theory offers novel advantages over existing approaches: 1) the relative positions of the chords in the time domain are retained, allowing the graphs to represent entire harmonic sequences of musical material, and 2) sequences from multiple sources are combined into a single graph, exposing features which are common to the source musical material, but which may vary or be absent from any particular instance. To test the schema, graphs were generated from recordings of ‘Georgia on My Mind’. We were able to produce examples demonstrating how this schema could be used to identify the essential harmonic framework of the song, to gain insight regarding the usage of chord substitutions by an artist during a single performance of the song, and to compare chord choices by two artists representing two different genres.