Correlational Harmonic Metrics: Bridging Computational and Human Notions of Musical Harmony

Abstract

The goal of this paper is to show that traditional music information retrieval tasks with well-chosen parameters perform similarly using computationally extracted chord annotations versus ground-truth annotations. Using a collection of Billboard songs from the last 60 years with provided ground-truth chord labels, I use established automatic chord identification algorithms to produce a corresponding extracted chord label dataset. I devise methods to compare chord progressions between two songs on the basis of their optimal localized alignment scores, adapting traditional sequence alignment techniques for transposition-invariance and the chord alphabet. I create a set of chord progression comparison parameters defined by chord distance metrics, gap costs, and normalization measures and run a black-box global optimization algorithm to stochastically search for the best parameter set to perform chordal comparisons on collections of songs across two primary tasks—fully connected harmonic comparison and query by n-grams. The first task involves evaluating chord progression similarity between all pairwise combinations of songs, separately ranking results for ground-truth and extracted chord labels, and returning the Spearman rho rank correlation coefficient of the two resulting rankings. The second task harmonically compares random chord query sequences of different sizes to the songs in the datasets, for each query ranking results for ground-truth and extracted chord labels, and returning the average Spearman rho rank correlation coefficient of all pairs of resulting rankings. These methods reflect common harmonic music information retrieval objectives and are robust and rapid, performing more efficiently than existing chord sequence alignment methods and introducing the use of correlational harmonic metrics between collections of songs.