Audio Darwinism

January 11, 2010 7:51 AM

DarwinTunes is an experiment in using genetic algorithms to create music.

DarwinTunes is a joint project between two professors of Imperial College in London. It is about to wrap up an experiment using Internet radio and ratings from the public to generate musical content. Volunteers rate a number of generated loops while listening. The best are used to create the next generation. (via slashdot)

The creators of DarwinTunes have previously been responsible for CompareStuff and Evolectronica (Google cache) and several television documentaries.

posted by mkb (13 comments total)

I have a haunting suspicion that this is the musical version of the Mechanical Turk, but it's Brian Eno hiding inside a small box somewhere inside the internet.
posted by chambers at 7:58 AM on January 11, 2010

Similar: Batuhan Bozkurt made a class for the SuperCollider programming language that given a synthesizer and a sound file uses a genetic algorithm to find the settings that most closely match the original sound.
posted by idiopath at 8:01 AM on January 11, 2010 [3 favorites]

chambers: "I have a haunting suspicion that this is the musical version of the Mechanical Turk, but it's Brian Eno hiding inside a small box somewhere inside the internet."

This is actually how Eno makes his music, but instead of anonymous volunteers judging the automatically produced output, it is Eno.
posted by idiopath at 8:03 AM on January 11, 2010 [1 favorite]

Wow, idiopath. I can't decide if I'm happy that the guy has done something I've thought up or angry that I have to find a new project.
posted by mkb at 8:06 AM on January 11, 2010

It sounds to me like no matter how much input comes from volunteers, the music is still going to be limited to the same jackhammer 16th notes.
posted by dunkadunc at 9:07 AM on January 11, 2010

I was thinking that these would make superior ringtones: seems like they've provided one a mechanism for downloading the loops, so maybe its possible...
posted by Ogre Lawless at 9:27 AM on January 11, 2010

Well, dunkadunc, there have only been 500 generations, and the only input from volunteers is "How good is this loop?" Depending on their data model, more esoteric constructions could arise. But then, they probably wouldn't survive long.
posted by mkb at 9:41 AM on January 11, 2010

Well, dunkadunc, there have only been 500 generations, and the only input from volunteers is "How good is this loop?" Depending on their data model, more esoteric constructions could arise. But then, they probably wouldn't survive long.

Yeah. These kind of genetic algorithms work well for optimizing something and coming up with clever solutions sometimes, but in the real world you have not only a fitness function but also competition between different animals.

Animals evolve in tandem and so you get all these ecological niches. Like Lions evolve sharp teeth to eat grazing animals, who then develop fast legs, or thick skin and horns to fight them off. A lot of people think human's big brains developed in part response to ever-increasing complex societies in monkey groups.

So really to get a lot diversity in the music here they need to figure out a way to turn different groups of listeners into different "ecological niches". It shouldn't be too hard though. They could try to cluster people based on the kinds of tracks they like.

Then of course they could try mixing (in genetic algorithm parlance, mating) songs from different niches. I doubt they would end up with genres that really match up with things like hip-hop, punk, heavy metal, but you could have different psudogenres/'ecological niches' based on different intervals, timing, etc.
posted by delmoi at 9:52 AM on January 11, 2010

Wow, idiopath. I can't decide if I'm happy that the guy has done something I've thought up or angry that I have to find a new project.

"Ooh, poop. Someone's already done oil painting."
It's not like academic research where the first guy to discover a new algorithm/particle wins.

Just roll your own GA with a different seed population and sound,
posted by sebastienbailard at 10:07 AM on January 11, 2010

sebastienbailard: "Just roll your own GA with a different seed population and sound,"

Actually this is pretty much academic stuff - the end user provides the seed and the sound and the synth, his class does the fitness selection and mutation. You could make a better one but it would be a one or the other sort of thing, it is more on the technology than the art side of synthesis.

The place for improvement on what he did would be in a better mutation algorithm or better fitness testing vs. the provided sound file.
posted by idiopath at 10:41 AM on January 11, 2010

The Autobiography of Charles Darwin (which is really a biography by his son) describes how Darwin enjoyed hearing music and had consistent tastes, but was unable to recognize tunes, even ones that he had heard many times and particularly enjoyed. When he heard a favorite tune, he would say something like "Say, that's nice! What is that?"
posted by neuron at 12:39 PM on January 11, 2010 [1 favorite]

I have a theory that it evolved into the upper-octave tinkles because most people have crappy laptop speakers and all they can hear are high frequencies. Or as my s/o puts it, because it sounds like winning money on slots at a casino.
posted by RobotVoodooPower at 5:15 PM on January 11, 2010 [1 favorite]

And now they're done! Here is the 500th generation.
posted by mkb at 6:10 AM on January 12, 2010