Rate me baby one more time: crowdsourcing the perfect song

July 13th, 2012 by

microtask_ratemebabyMusic is a powerful thing. It can make you laugh or cry, and even tear down the Berlin Wall (according to David Hasselhoff, anyway). The same song can mean different things to different people, with infectious tracks like Britney Spears’ ‘Hit Me Baby One More Time’ managing to be both the all-time most requested song on US radio as well as an instrument of torture used at Guantanamo Bay (along with ‘I Love You,’ the theme from Barney the Dinosaur, which would make me crack within seconds).

Thanks to the internet, we now have access to more music than ever before, and unless we’re trapped in a military prison, we can listen to whatever we want. iTunes lists a bewildering array of genres, from Neo-Medieval Disco to Japanoise to Pirate Metal. With so much variety on offer, and so many different types of people, is it possible to create a song that everybody will love?

A natural selection of your favorite tracks

A group of evolutionary scientists at Imperial College in London have created a crowdsourced research project which attempts to do just that. The researchers note that every time a listener chooses one song over another, they are making a creative choice, with one track being more successful than another. This process mirrors Darwinian natural selection and provides a great way to test the idea that music can evolve and adapt.

The project’s central experiment began with the creation of a piece of software called DarwinTunes, an automated music-creation system. It makes new songs from sound fragments and loops, and requires listeners to rate each song on a 5-point scale from ‘I can’t stand it!’ to ‘I love it!’ Successful loops and sounds are incorporated into the next generation of tracks, while unpopular ones are phased out, like the musical equivalent of pedigree dog breeding. A 7,000-strong crowd formed the judging panel for the initial experiments, and has so far taken the music through over 2,500 generations.

The first generation of loops was randomly generated, and barely sounded like music at all. But as the crowd selected loops which contained pleasing chords or sequences of notes, tunes began to emerge. Soon, the randomly generated sounds had become music, with popular melodies or chords surviving through the generations and cross-breeding with other successful loops.

The sound of the crowd

But what does this crowd-guided music actually sound like? At first listen, the hypnotic cascades of melodic bleeps and bloops reminded me of an early ’90s role-playing game, filling me with an irresistible urge to drink potions and rescue a princess. Though all DarwinTunes’ loops share the same palette of sounds, in order to keep the focus on melody there are a variety of distinctive tunes competing for attention. Some are sad, some cheerful, and some are downright catchy (Loop 81 in particular caught my ear, but that may be because it bears a slight resemblance to the greatest piece of video game music ever sequenced).

In the real world certain fragments of melodic DNA do seem to crop up again and again, but it’s hard to imagine legions of screaming fans queuing overnight for the chance to hear Loop 81 performed live. However, the researchers from Carnegie Mellon are not only interested in creating the perfect song. The project also provides useful data on the way individual taste and preference affects group decision making. This is a key area for future crowd research, as crowds grow larger and incorporate members from many different cultures.

More and more research is revealing the delicate balance at the heart of the “wisdom of crowds” effect: crowds thrive on diversity, but there needs to be common understanding too. By exploring our similarities, this kind of research may help us to understand and overcome our differences. Even if it doesn’t lead to the greatest song ever written, the DarwinTunes project will have a lot to tell us about what we have in common (plus, even the worst loop is bound to be better than Barney the Dinosaur, and that has to count as a victory).

  • http://twitter.com/Huopainen MoreLiver
  • Evan Duxbury

    Cool article Ville. Hype Machine attempts to do something like this with their favouriting system. Not quite as scientific, but it does go a long way to filtering out the noise and gives you the whole song rather than just a teaser. Where do you draw the line when it comes to crowdsourcing? Does Hype Machine count?