The data visualists over at The Pudding have put together a fascinating analysis of pop music going back to the 1950s using an eight-point measure originally developed at MIT.
From the article:
In 2005, Tristan Jehan, a PhD student at MIT, published his dissertation, “Creating Music By Listening,” a framework for AI-created music. The immediate implication of Jehan’s work has not been the human-to-robot gift of art, but reducing a song to a small set of data points that say something about the song generally, such as “valence” (roughly the happy to sad spectrum) and “energy” (arousing to soporific). Shortly after completing his dissertation, Jehan co-founded a company called EchoNest, and the data became a pillar of Spotify’s recommendation systems, determining music similarity and accurately suggesting songs that sound alike.
This data is publicly available, and to measure whether songs sound similar, we’ll calculate the differences in EchoNest’s 8 data points for top songs in the Billboard Hot 100, a peer-reviewed method employed by other music researchers. In theory, the songs with most similar EchoNest values should sound similar as well.
The result is a trend toward similarity, with smaller distances among songs. To date, songs that charted between 2012 and 2016 were the most similar, according to EchoNest data.
John Seabrook is a staff writer at the New Yorker and author of The Song Machine, a history of pop music’s last 25 years of unstoppable hit-making sophistication. When he looked at this trend, what did he see?
“What I see,” he said, “is the enormous, quick but large scale shift from the kind of craft-like song-making process of people putting together lyrics and melody in a semi-organic setting to what I call the track-and-hook method…”
For the record, I like pop music, but this does seem to confirm my suspicion that it’s more canned and commercial than ever before. There’s nothing wrong with enjoying a piece of candy. You should just know that’s what you’re eating.