On Wandering the Paths of a Spotify Analyst’s Mad Music Map

Have you ever wondered how many music genres there are? Spotify data analyst Glenn McDonald has an answer to the question — and he’s built the map to illustrate it. Every Noise at Once catalogues over 1,500 genres ranging from the esoteric (modern uplift, deep discofox, power violence) to the outright bizarre (solipsynthm, terrorcore, and something called catstep). Clicking a style triggers its audio sample. A tiny arrow near the genre’s name leads to a smaller map collecting its artists. Bigger names are, well, bigger names. Choosing an artist leads to the corresponding Spotify artist page. Voilà! Instant music discovery. Search the main map by the genre or artist of your choice (who knew Radiohead were classified as melancholia?). Clicked genres feature a music note icon for remembering favorites.

The curious map arranges Patton Oswalt’s acerbic comedy near the fizzling purity of the delta blues. Anime score’s swirling strings are neighbors with freak folk and football anthems. Every Noise’s curious constellation fascinates even the casual listener, but its data madness will also produce puzzled looks. According to McDonald, the map relies on 13 variables to cluster similar genres and artists. “The thirteen audio dimensions include tempo, loudness, energy, emotional valence and danceability,” he says. “The vertical axis represents the spectrum from mechanistic to organic.” Genres with rigid tempos and electronic instrumentation like tech house, minimal dub, and the peppy style of bubble trance are at the top of the page. “The bottom contains organic music, which combines acoustic instrumentation and variable human timing. They contain more of a human element.” Classical piano, harp music, and expressive genres less concerned with metronomic timing live here.

The horizontal axis represents sonic density. On the left is the roaring crunch of cryptic black metal, but also classical organ, which is sonically dense but not as noisy. On the right, music is “bouncier”, consisting of sharper noises and clear spaces between notes. Roots reggae, Turkish hip-hop, and less rhythmic, sparser styles like poetry reading fit the mold.

The microgenre explosion in the 21st century, aided by a combination of software advances, faster internet connections, and the globalized proliferation of music, has provided listeners with a comprehensive selection of styles, all of which are catalogued on Every Noise. McDonald prefers to think of genres as fuzzy, fluid. “I’m cheerfully unrigorous about what a genre can be. Some of them are musicological, some are historical, and others are regional themes. A genre can be any listening mode I think people might want.” The map’s development was an experiment. “It all came about while I was looking for ways to crosscheck acoustic attributes. Aggregating them at the genre level helped determine whether they were working.” McDonald realized some layouts were more visually interesting than others and settled on the scatter plot to represent the data. The color system also arose from a trial-and-error process to map three more of the 13 attributes using shades of red, green and blue. “The colors indicate dissimilar neighbors. For example, you can follow the blue-purple hues across the middle of the map to track a bunch of ambient, atmospheric genres that have different positions in the two-dimensional space, but share a timbral quality. The same organizational systems and colors are applied to the artist pages.”

The map provides a new music excavating method, helping us find similar bands to Of Montreal (they’re identified as “indie Christmas”) based on their aggregation. Bands can pursue the scarce gaps in the topology to ensure they’ll sound distinct from their peers.

The map goes deeper. When inside a genre, there are two inset maps under the main artist map. The first shows the genres closest to the one you’re inside. Culturally-related genres share the same artists, as in Croatian pop and Croatian hip-hop. While not exactly filtering, it’s a way to look at the data on a finer level and not get lost in the clutter of the main map.

The second inset map, in black, shows the styles furthest from the current genre. McDonald is eager to relate its creation. “This is a funny story. It was my goal to say, ‘if you hate this style of music, then what’s the farthest thing from it?’ as presumably, you’d like that. But when I programmed it, I found the answer was always satanic black metal, esoteric tech house, classical piano or political speeches. Those four points were the extremes, and no matter where you were, one of those was the farthest away. That wasn’t interesting. Months later it occurred to me to try to interpret the data in a different way to achieve a true inverse. So if your genre is slightly faster than average, the black map picks something slightly slower, with something that’s equally conservative or extreme, but at the opposite polarity in each dimension. That turned out to be much more interesting.” You can now spend hours marveling at how appropriate it is that trap’s opposite is kraut rock.

A number of genres are designated “deep”. Classical piano contains pianists most people are familiar with, while deep classical piano might collect artists who have won obscure piano competitions. Musically the genres could be virtually identical, as in classical, which relies on performing the canon. Deeper genres tend to correlate with different listeners and lesser plays. “In some cases, there are subtle but identifiable distinguishing characteristics. A particular flavor of tech house will actually tend to produce slightly different music. It will adhere more rigorously to some template. Its playlist will have a more consistent feel than one that’s closer to the surface, where more influences and cross-genre stuff happens.” Dive into the deep if you’re tired of the artists everybody knows about. Satisfy your inner hipster.

Genres are started, or “seeded”, by identifying a set of artists. Creedence Clearwater Revival, The Who and Lynyrd Skynyrd would be a good seed for classic rock. The automation and data analysis that chooses other classic rock bands is a combination of how people listen, what the music sounds like, and what people say about it online. “We scrape millions of online sources a day, analyzing the text to figure out who it’s about and what people are saying. The procedure figures out the words people use to describe a given artist.” This isn’t a perfect process, however. “Metal” might be associated with someone not currently in a metal band, but the terms describing most metal bands tend to be similar.

Even understanding Every Noise’s functionality, the site has its limitations. For one, there’s no way to sort the map, so it’s not possible to see how many electronic genres have arisen since 2010, for example. However, there are some links to other data views at the bottom of the map. One of these, The Retromatic History of Music (or Love), maps genres across time. “It’s a little mushy because the release dates we have are accurate generally, but subject to a lot of individual error.” McDonald tried ranking the top songs of 1953, then realized 70 of the top 100 gathered were released in a different year. This is an intermittent problem with Spotify, which is where Every Noise gets its data. For example, industrial artist ohGr’s 2001 debut Welt shows up as being released in 2003, coming ahead of 2003’s Sunnypsyop, an inaccurate chronology. “In almost any year, you’ll have individual errors. They don’t invalidate the overall analysis, but they make it unsatisfying to make these types of playlists and charts.”

The most common error is artist-name ambiguity. “If there were a global band-name registry, it would really help everybody. Absent that, we just have to fix stuff.” A search for “Lee Jones” brings up the DJ and the singer-songwriter incorrectly combined into the same artist. “There’s potential human involvement at almost any point, but we try to intervene in the least tedious ways. If a genre requires too much human oversight, we reconceive it as something to which the data lends itself more naturally.”

Limitations aside, the map provides heaps of useful data, like which genres are the most popular. “Obviously hip-hop is huge and constantly evolving everywhere. Nearly every country has something interesting happening around hip-hop. Reggaeton-influenced versions are particularly hot in Latin America and southern Europe. Pop punk and second/third-generation emo are big with younger listeners now.” Seems like teenage angst is universal, no matter the generation. The Every Noise list linked here analyzes youth demographics, the source of all music trends.

Though McDonald’s day job at Spotify means intermittent work on Every Noise, the site’s analysis is at the core of Spotify’s analytics, determining how features appeal to different audiences. A recently implemented feature is the Daily Mix, functioning as a listener’s personal genre. Some data techniques from Every Noise have been ported over to the popular listening app. “I’m centrally or peripherally involved in a lot of things. I meddle in stuff! ‘The Sound of‘ playlists are entirely my end work on top of layers and layers of hundreds of other people. I also work on things feeding into the Related Artists. Every Noise is actually a fairly thin visualization layer over the categorization and similarity data, which gets used at different layers inside Spotify.”

The mind behind the map

McDonald admits he hasn’t added any new genres in a while, but will soon do another round. “Some of the underlying pieces got upgraded to better data sources and ideas. Every time we improve some part of the process, I identify things I hadn’t before. Sometimes, it consolidates or splits things. It increases the resolution, if you will.” Greater granularity will enable the data to identify new music scenes. “Ideally, I’d like to be able to distinguish between Lithuanian pop and Lithuanian hip-hop, but I can’t because there just isn’t enough of it. In some cases there’s plenty of music but I don’t have enough data to tease them apart. But each time we improve something, that becomes more possible.” That possibility is what musical freedom is all about.