Abstract
This paper compares two corpora of melodies drawn from premillennial and postmillennial American popular music, and identifies several notable differences in their use of rhythm. The premillennial corpus contains melodies written between 1957 and 1997 [deClercq and Temperley (2011. “A Corpus Analysis of Rock Harmony.” Popular Music 30 (1): 47–70)], while the postmillennial corpus (compiled for this study) consists of songs popular between 2015 and 2019. For both corpora, we analysed (1) the distribution of note onsets within measures; (2) the distribution of four-note rhythmic cells, (3) the speed of melodic delivery, and (4) the tempo of the tactus. Our analyses indicated that the postmillennial melodies are delivered more quickly, are distributed more evenly throughout their measures, repeat rhythmic cells more frequently, and are annotated at slower tempos. Even when the tactus tempos were standardized into an allowable window of 70–140 BPM, this effect, though smaller, remained. We then use our techniques to observe the properties of three representative postmillennial tracks, finding that salient information can be located in both standardized and non-standardized tactus data, and that tempo-variant differences between corpora are closely connected to musical genre, with music designated as “pop” being more similar over both genres, and postmillennial rap and hip-hop introducing the most uniqueness.
2012 Computing Classification Scheme:
Acknowledgements
We would like to thank Trevor de Clercq, the anonymous reviewer, and Jason Yust for their guidance on the preparation of these materials.
Disclosure statement
No potential conflict of interest was reported by the authors.
Supplemental online material
Supplemental online material for this article can be accessed at chriswmwhite.com/popannotations
Notes
1 There are arguments in favour of a number of reporting methods in the corpus analysis, machine learning, and popular-music analysis literature. Importantly, the goal of this study is to describe the differences between two corpora, not to claim experimental replicability (the general motivation behind confidence intervals). Throughout this paper, we therefore use a number of methods to describe variance, not with the motivation to claim replicable difference but to describe the difference between two datasets in their manifested versions and suggest provocative corpus properties which may inspire future studies.
2 This is an interesting reversal of the trend noted in CitationTan, Lustig, and Temperley (Citation2018), in which songs seem to use fewer offbeat sixteenth notes in the later decades in their corpus; they speculate this is because of an increase in average tempo in the later twentieth century. However, this quickening of pulse in popular song is also noted in recent hip-hop music in CitationDuinker (2020a). Additionally, the increase in quicker note values tracks with the centuries-long trend associated with quickening metric values in Medieval and Renaissance vocal music identified in CitationDeFord (Citation2015). Additionally, it should be noted that CitationTan, Lustig, and Temperley (Citation2018) mark the last eighth-note in the measure as a frequent position for syncopation in their analysis of premillennial popular melodies, something evidenced by the relatively high probability mass at that point in our premillennial metric profiles; given that our analysis shows a significant decrease in attacks at that point in the postmillennial profiles, it would seem that this tendency may disappear in postmillennial popular melodies.