¶ Metropopolis · 4 December 2014 listen/tech
Spotify just published the super-cool #YearInMusic thing, which shows a variety of statistical excerpts and summaries of both Spotify global listening and (if you're a Premium subscriber) your own for 2014.
Among other things, the global feature includes an editorial citation (which I had nothing to do with picking) for 2014's "Breakout Genre", Metropopolis.
You might not have heard of Metropopolis. A few people who also hadn't heard of it wrote indignant articles about this fact:
- Here Is Spotify's List of the Most Streamed Music of 2014, or What the Fuck Is Metropopolis? (from Noisey/Vice)
- Stop trying to make "metropopolis" happen: How Spotify forged a dubious new musical genre (from Salon)
Conversely, hundreds of people who hadn't heard of it either have now subscribed to the playlist that demonstrates what it means.
It doesn't actually much matter what it's called, I think. It's not a breakout genre name, it's a set of music that was breaking out in 2014. If you don't know what the name means, click the link and listen to it, and then you'll know. I don't care if it "happens", I care that you find more music you might love, and this might be some of it.
But it does also have a name, and I happen to know the story of how there came to be a thing with this name.
One of the things I watch over at work is the Echo Nest / Spotify list of genres. A genre can be any of many kinds of things, and can be varying degrees of known or unknown. In practical terms, "genre" for us kind of just means "thematic listening cluster", and our goal as we have expanded the list is to find and name and track as many such thematic listening clusters as we can identify in the world.
In most cases, these clusters exist in the world with a name. "Album rock" is a thing. "Samba" and "Nintendocore" and "Sega" are things (and Sega has nothing to do with Nintendocore!). We can model and track these, and make playlists to express them, but we don't have to name them.
But not all the clusters we find come with already-culturally-established names. What do you call the emerging cluster of loosely r&b-derived, often synthpop-orchestrated, generally sensual music that people like Frank Ocean and How to Dress Well are making? Various music-critics have suggested "pbr&b", "hipster r&b" and "r-neg-b", but most people who like the kind of music we're talking about don't know those terms. We originally named this cluster "r-neg-b", because of those three that was the one with the lowest amount of smug derision. But people kept accusing us of making that up, so we recently switched to calling it "indie r&b", on the theory that the music is kind of a cross between indie pop/rock/folk and r&b, and at least maybe you can guess what "indie r&b" might mean, whereas "r-neg-b" reads like a character-encoding error.
Or sometimes multiple clusters come with the same name. When you say "trap", for example, do you mean trap or trap? One is a hip-hop subgenre, the other is a largely-instrumental electronic subgenre. For our list, then, we have opted to eliminate the ambiguity by calling the hip-hop one "trap music" and the electronic one "trapstep". There's nothing magically right about those names, but they are at least different from each other.
And then, sometimes, because at Spotify we have maybe more data about human music-listening patterns than has ever existed before, we find clusters that you otherwise probably wouldn't be able to isolate, and thus wouldn't even have thought to name. For example, you know those rousing neo-rustic folk/pop-ish artists like Mumford & Sons and the Lumineers that kind of sound like Dave Matthews ran over a jug band? We have the data to make a listening cluster around those with hundreds of other someway-similar bands. If you like that kind of music, you'll probably enjoy listening to this cluster. But it doesn't come with a name, exactly. So we called it "stomp and holler". We didn't totally make up that term, but it wasn't a genre name before we said so.
And you could argue that it still isn't a genre name, really, after we said so. This is fine. We don't have a philosophical or taxonomical agenda, we just have these clusters of awesome (usually) (to somebody) music, and we need some words to use as labels on a map, or as titles on playlists. When I have to make up names, I try to do so using the absolute minimum amount of creativity necessary to produce a unique new phrase, and thus we get a lot of rather mundane coinages like Malaysian pop and traditional reggae and atmospheric black metal. Sometimes I resolve name-ambiguity by the innovative linguistic wizardry of adding the words "more" or "deep", and thus we get a series of methodical techno clusters called deep house, deeper house and more deeper house. On the one hand, these are dopey names. On the other hand, if you like that kind of music, I'm betting that, in the same way that I continue to listen to BUMP OF CHICKEN, you'll still like listening to it long after you get over the name. (Or embrace it.)
Every once in a while, though, I lack the imagination to think of a boring name, and am thus forced to settle for a creative one. This is how the cluster of theatrical melodic metal with mostly operatic female vocals came to be called fallen angel. This is how the cluster of music that can sometimes sound like people singing distractedly while dissolving parchment sheet-music in beakers of gurgling solvent came to be called laboratorio. This is how the cluster of music that used to be New Wave only we're still listening to it now that it's old came to be called permanent wave. This is how we came to have shimmer pop and shiver pop and soul flow. I'd pick duller names if I could, but the names just exist to get you to the music.
The music, in all cases, is actually picked by computer programs using math to distill massive quantities of data. No matter what label I apply, these clusters exist because the world of people who make, listen to and write about music has collaboratively brought them into being by playing and listening and writing in particular combinations of patterns. In most cases, the computer programs use all this data to do two things: first they try to pick a set of cluster-appropriate artists, and then they try to pick those artists' most cluster-appropriate songs.
This often works, but not always. Take, for example, piano rock. The numbers we calculate to characterize songs don't identify individual instruments, so if you let the computers pick artists that fit the "piano rock" mold, you get a bunch of rock with pianos, but also a bunch of similar rock by bands that don't actually ever use pianos. We could have let this happen, and renamed the cluster "post-maudlin rock", but in the spirit of avoiding smug derision, we instead went through the artist-list by hand and made it deliver, at least roughly, on the promise of "piano rock".
And this is how we got metropopolis, too. I was listening, at one point, to a lot of indietronica, but when the computers made their indietronica playlist, I found that about half of it sounded like Chairlift and Chvrches to me, but half of it didn't. Which wasn't a problem, because "indietronica" doesn't have to sound like Chairlift and Chvrches, it just has to sound indie and tronic. But I wanted the cluster that did sound like Chairlift and Chvrches. So I made it. I had some other candidate names that I have since forgotten, but "metropopolis" seemed obviously better than the others as soon as it occurred to me, some kind of shiny aesthetic futurism with an insidious dystopian undertone.
I watch over this cluster myself. The computers are actually pretty good at suggesting potential additions, but I take the time to go through and listen to each one, and only put them into the cluster if they sound sufficiently metropopolistic. This is, from my point of view, an admission of temporary defeat. The computers ought to be able to do this by themselves. If we had a few more dimensions of audio analysis, quantifying just a few more psychoacoustic attributes, maybe we could isolate the precise buoyant glitteriness I hear, or the kind of resigned muting of energy that distinguishes some of the data candidates I reject. I don't, ultimately, think this cluster is any different from liquid funk or doo wop. It's a thing, I can hear it. The computers can't hear it yet. And I wanted to listen to it more than I wanted to wait for them to learn.
Among other things, the global feature includes an editorial citation (which I had nothing to do with picking) for 2014's "Breakout Genre", Metropopolis.
You might not have heard of Metropopolis. A few people who also hadn't heard of it wrote indignant articles about this fact:
- Here Is Spotify's List of the Most Streamed Music of 2014, or What the Fuck Is Metropopolis? (from Noisey/Vice)
- Stop trying to make "metropopolis" happen: How Spotify forged a dubious new musical genre (from Salon)
Conversely, hundreds of people who hadn't heard of it either have now subscribed to the playlist that demonstrates what it means.
It doesn't actually much matter what it's called, I think. It's not a breakout genre name, it's a set of music that was breaking out in 2014. If you don't know what the name means, click the link and listen to it, and then you'll know. I don't care if it "happens", I care that you find more music you might love, and this might be some of it.
But it does also have a name, and I happen to know the story of how there came to be a thing with this name.
One of the things I watch over at work is the Echo Nest / Spotify list of genres. A genre can be any of many kinds of things, and can be varying degrees of known or unknown. In practical terms, "genre" for us kind of just means "thematic listening cluster", and our goal as we have expanded the list is to find and name and track as many such thematic listening clusters as we can identify in the world.
In most cases, these clusters exist in the world with a name. "Album rock" is a thing. "Samba" and "Nintendocore" and "Sega" are things (and Sega has nothing to do with Nintendocore!). We can model and track these, and make playlists to express them, but we don't have to name them.
But not all the clusters we find come with already-culturally-established names. What do you call the emerging cluster of loosely r&b-derived, often synthpop-orchestrated, generally sensual music that people like Frank Ocean and How to Dress Well are making? Various music-critics have suggested "pbr&b", "hipster r&b" and "r-neg-b", but most people who like the kind of music we're talking about don't know those terms. We originally named this cluster "r-neg-b", because of those three that was the one with the lowest amount of smug derision. But people kept accusing us of making that up, so we recently switched to calling it "indie r&b", on the theory that the music is kind of a cross between indie pop/rock/folk and r&b, and at least maybe you can guess what "indie r&b" might mean, whereas "r-neg-b" reads like a character-encoding error.
Or sometimes multiple clusters come with the same name. When you say "trap", for example, do you mean trap or trap? One is a hip-hop subgenre, the other is a largely-instrumental electronic subgenre. For our list, then, we have opted to eliminate the ambiguity by calling the hip-hop one "trap music" and the electronic one "trapstep". There's nothing magically right about those names, but they are at least different from each other.
And then, sometimes, because at Spotify we have maybe more data about human music-listening patterns than has ever existed before, we find clusters that you otherwise probably wouldn't be able to isolate, and thus wouldn't even have thought to name. For example, you know those rousing neo-rustic folk/pop-ish artists like Mumford & Sons and the Lumineers that kind of sound like Dave Matthews ran over a jug band? We have the data to make a listening cluster around those with hundreds of other someway-similar bands. If you like that kind of music, you'll probably enjoy listening to this cluster. But it doesn't come with a name, exactly. So we called it "stomp and holler". We didn't totally make up that term, but it wasn't a genre name before we said so.
And you could argue that it still isn't a genre name, really, after we said so. This is fine. We don't have a philosophical or taxonomical agenda, we just have these clusters of awesome (usually) (to somebody) music, and we need some words to use as labels on a map, or as titles on playlists. When I have to make up names, I try to do so using the absolute minimum amount of creativity necessary to produce a unique new phrase, and thus we get a lot of rather mundane coinages like Malaysian pop and traditional reggae and atmospheric black metal. Sometimes I resolve name-ambiguity by the innovative linguistic wizardry of adding the words "more" or "deep", and thus we get a series of methodical techno clusters called deep house, deeper house and more deeper house. On the one hand, these are dopey names. On the other hand, if you like that kind of music, I'm betting that, in the same way that I continue to listen to BUMP OF CHICKEN, you'll still like listening to it long after you get over the name. (Or embrace it.)
Every once in a while, though, I lack the imagination to think of a boring name, and am thus forced to settle for a creative one. This is how the cluster of theatrical melodic metal with mostly operatic female vocals came to be called fallen angel. This is how the cluster of music that can sometimes sound like people singing distractedly while dissolving parchment sheet-music in beakers of gurgling solvent came to be called laboratorio. This is how the cluster of music that used to be New Wave only we're still listening to it now that it's old came to be called permanent wave. This is how we came to have shimmer pop and shiver pop and soul flow. I'd pick duller names if I could, but the names just exist to get you to the music.
The music, in all cases, is actually picked by computer programs using math to distill massive quantities of data. No matter what label I apply, these clusters exist because the world of people who make, listen to and write about music has collaboratively brought them into being by playing and listening and writing in particular combinations of patterns. In most cases, the computer programs use all this data to do two things: first they try to pick a set of cluster-appropriate artists, and then they try to pick those artists' most cluster-appropriate songs.
This often works, but not always. Take, for example, piano rock. The numbers we calculate to characterize songs don't identify individual instruments, so if you let the computers pick artists that fit the "piano rock" mold, you get a bunch of rock with pianos, but also a bunch of similar rock by bands that don't actually ever use pianos. We could have let this happen, and renamed the cluster "post-maudlin rock", but in the spirit of avoiding smug derision, we instead went through the artist-list by hand and made it deliver, at least roughly, on the promise of "piano rock".
And this is how we got metropopolis, too. I was listening, at one point, to a lot of indietronica, but when the computers made their indietronica playlist, I found that about half of it sounded like Chairlift and Chvrches to me, but half of it didn't. Which wasn't a problem, because "indietronica" doesn't have to sound like Chairlift and Chvrches, it just has to sound indie and tronic. But I wanted the cluster that did sound like Chairlift and Chvrches. So I made it. I had some other candidate names that I have since forgotten, but "metropopolis" seemed obviously better than the others as soon as it occurred to me, some kind of shiny aesthetic futurism with an insidious dystopian undertone.
I watch over this cluster myself. The computers are actually pretty good at suggesting potential additions, but I take the time to go through and listen to each one, and only put them into the cluster if they sound sufficiently metropopolistic. This is, from my point of view, an admission of temporary defeat. The computers ought to be able to do this by themselves. If we had a few more dimensions of audio analysis, quantifying just a few more psychoacoustic attributes, maybe we could isolate the precise buoyant glitteriness I hear, or the kind of resigned muting of energy that distinguishes some of the data candidates I reject. I don't, ultimately, think this cluster is any different from liquid funk or doo wop. It's a thing, I can hear it. The computers can't hear it yet. And I wanted to listen to it more than I wanted to wait for them to learn.