Quantcast

This Guy Has Mapped Music, and Now He Wants to Map Your Mind

Travelling through genres has never been such smooth sailing.

Dan Wilkinson

The idea of a musical map works for me. Music genres are incesteous; they are connected in more ways than the M25 and having a guide map helps piece together the junctions. For example: a minor shift in tempo and early R&B and reggae start to share similarities. The same goes for punk and nu-wave, hip-hop and trap, or pop and whatever PC Music are making.

Echo Nest are a company who calculate and compile your listening habits on Spotify (including that one time you listened to DJ Snake for an hour straight). They wanted to find an exact technique for the genres we connect artists with - and they devised a musical map (called Every Noise at Once, and pictured below) linking every recorded genre that’s been invented. The result: one of the most comprehensive music genre lists out there. The future fuel to many ill advised new-age playlists.

I spoke to Glenn Mcdonald, the principal engineer at Echo Nest about making the map, having too much musical power, and what kind of big brother Spotify really wants to be.

Every Noise at Once

Noisey: How did Every Noise at Once come about?
Glenn: The underlying genre project came about because people would ask the Echo Nest to play them some rock music - we would play Rihanna, and that was obviously not what anybody meant.

Why Rihanna for rock?
We knew what descriptive terms were relevant to any given artist, and "rock" is one of many relevant terms people sometimes use in talking about Rihanna. We now needed to know the reverse: which artists are most relevant to a given term?

How did you find the genres originally?
I started with a long list of 500 or so commonly used descriptive words or phrases, all from our ongoing web crawling of millions of sources of musical information and stuff online. Maybe 300-ish of these were actually genres - as opposed to places or instruments or non-genre terms like "loud" or "beautiful" or whatever. At first there were a lot of obvious additions I could think of just from general knowledge or small amounts of research.

Like what?
In later waves I got into automated mining of the data-space to discover clusters of related artists - for which we didn't have genre-labels. Then I would have to go figure out what the right labels were.

How do you mean?
I mean, at first it was as easy as "Let's see, do we have 'skiffle'? I know that's a thing. No? OK, we'd add that. Next?" In the mining stage - where I’d pool all the data together - I'd get a bunch of auto-clustered artists, and a list of weighted terms to describe them. Then I'd have to look at the terms and realise that, say, "fingerstyle" was actually a genre name. Then, in a small but entertaining number of cases, I had to make up a "new" name.

What names did you create?
Most of these are just functional; like "traditional rockabilly", or "atmospheric black metal". There are a handful of more creative ones I just made up; my favourites include "fallen angel" and "laboratorio". I try to only do this when there are no good existing options. But hey, somebody has to name these things.

How creative. What genre was the biggest in terms of the most artists covered?
Rock. But I guess that’s a given.

I guess guitar music isn’t dead on Spotify then. What about the genre for most subgenres?
The metal and dance communities are pretty big on naming little niches and movements.

What do you think of small scenes that don't get enough time to be a genre? Like chillwave, seapunk and vaporwave?
Well, I do have chillwave and vaporwave on the musical map. They may not be as cleanly delineated as some others, but I think they still produce listening experiences with a discernible character. If I can't find (or the math can't find) at least 100 artists who could be working in that genre, in some loose sense, then I don't include it. I tolerate a lot of fuzziness. So, chillwave? Sure. Pirate metal, where there's kind of more like 8 bands rather than 100? No. So, not crabcore, not Paisley underground, not Tromso techno, basically.



Is it scary having so much musical power?
Ha. No, but it definitely helped that I did most of the work on the genre space before there was a whole lot of attention being paid to it. I expect things to change a bit once the whole thing begins taking more visible shape inside of Spotify. Still, "power" is probably not quite the right word for the situation. Nobody's fortune is going to be made or lost based on whether they made it into my "power-pop punk" genre.

Tell that to Bowling for Soup. Doesn’t this all turn into genre politics?
I guess, but remember that most of the time it's not me you're arguing with, it's the emerging listening and music-discussion behaviour of the entire web. I mainly just channel it.

Do you think reworking the way people choose to listen to music (such as the map) is the key to getting them to listen and discover more?
It depends so much on the person. For some people, like me, that map is like a better dream future than jetpacks. Yet for some people the map is a brain-freezing nightmare of unmanageable chaos. Which is why I added the ‘List’ mode, which is a much simpler presentation, visually. Even so, some people don't have the time or temperament to explore that way. For those people we need to take a different approach, and give them something more like a guide than a map.

What do you feel companies like Next Big Sound (https://www.nextbigsound.com/) are adding to the way we consume music?
Listening data is amazing and very informative.

But where do you think it's all heading to in the future?
We've gone, incredibly quickly, from the days of radio and record-selling where the listeners were essentially anonymous, to the streaming world where we obviously know every track that every user plays on Spotify. That's just how it works. Although making useful sense out of all that data is way harder than it might seem. We know way more about what people do than why. Did you skip a song because you hated it, or because you love it and already played it 1500 times on CD? So personally, I approach these things with caution. Eventually, I think more data will make it possible to make better connections between artists and their listeners. As well as between listeners and their artists, depending on your point of view.

Will Spotify eventually become big brother?
Hopefully the big brother whose record-collection you borrowed.

Touché. What other information are you looking to process next?
I'd love to know more about people's intents when listening to music. Not invasively, but by giving people better tools for telling us what they're trying to accomplish choosing a particular song. Right now we usually have to guess, or rely on playlist names or things like that. Did you play that Motown list because you were sad and it cheers you up, or were you having a dinner party? What music helps you study? What do you like to run or work or nap or whatever to? One person's deep-focus music is another person's intolerable annoyance. I hate guessing and trying to read minds. I'd much rather let people just tell us.

So everything is all connected?
Definitely. Everything is always connected.

Thanks Glenn!