Advertisement
YOU ARE HERE: LAT HomeCollections

DarwinTunes software 'evolves' music without composers

Noise filtered through generations of listeners eventually becomes melodic, offering insights into how our tastes evolved.

June 19, 2012|By Rosie Mestel, Los Angeles Times
  • A screenshot from DarwinTunes, software that creates 8-second collections of notes and puts them through the evolutionary wringer.
A screenshot from DarwinTunes, software that creates 8-second collections… (Screenshot )

Composers, look to your laurels: A mere computer program can transform a racket of clangs, hums and beeps into a pleasing melody, and all humans have to do is offer feedback with the click of a mouse.

The program, by a British bioinformatics expert whose day job involves tackling biomedical problems, employs the same principles of natural selection that guide the evolution of living beings over many generations. The software — dubbed DarwinTunes, of course — creates 8-second collections of notes and puts them through the evolutionary wringer.

All the tunes have certain fixed traits: the same four-beats-per-measure tempo and tones from the Western 12-tone musical scale.

But other traits vary. Any note or instrument can be played at any given point at any volume and with any special effect. The result is a huge number of possible sound experiences, from the intolerable to the sublime.

Robert MacCallum of Imperial College London and his colleagues asked 120 undergraduates to listen to continuous loops of the computer-generated tunes and rate them on a 5-point scale that ranged from "I can't stand it" to "I love it." The tunes were also posted on the Web (at darwintunes.org/evolve-music), where anonymous critics added their feedback.

Each time tunes were rated, the half that came out on top were selected to contribute to the next generation of music. Each successful melody would mate with another successful melody, spawning similar-but-not-identical daughters before expiring from the collection.

Tunes that didn't make the cut were relegated to the evolutionary trash heap without leaving behind any musical legacy.

"All that matters is that some loops are better than others — that some are slightly less of a racket," said MacCallum, who has a long-standing interest in both music and computers. "I'm pretty sure that the painful screechy loops were eliminated first."

The process is slow, "a bit like listening to paint dry," he added.

But as the undergraduate experiment went through its 400 generations and the online experiment 2,500 generations, pleasing melodies began to emerge, MacCallum's team reported Monday in the journal Proceedings of the National Academy of Sciences. MacCallum noticed two main traits that the raters had apparently selected: more chords and more complex rhythms.

Past a certain point, though, the tunes in both experiments stopped evolving — the daughter tunes are less well liked than their parents, leaving the music stranded on an evolutionary plateau, the scientists wrote.

The approach, known as evolutionary computation, has been used to tackle a broad array of problems in such fields as theoretical biology, engineering and medicine — be it the compact design of a satellite antenna or development of a drug that will precisely lock onto its target.

In the realm of music, MacCallum said he didn't think human composers should feel threatened by DarwinTunes.

"It took 10 1/2 days of human time to create a four-bar Steve Reichian jingle," he said.

Still, the study offers interesting insights into how music has evolved through time, he said. Scientists know that there are rules for what constitutes a pleasing sound, some of which are culturally driven and some of which may be hard-wired in our brains. The results of this experiment suggest that listeners aren't just passive consumers but play a highly influential role in music's evolution.

This is not exactly a surprise, said composer David Cope, a professor emeritus of music at UC Santa Cruz who wasn't involved in MacCallum's study. Cope has created computer algorithms to analyze patterns of phrase and note choice in works by Bach, Beethoven and Chopin and used the rules to spawn "new" works by these long-dead masters. Others have written programs to compose jazz solos or predict hit tunes.

"Composers, like everyone else in the world, no matter how individualistic or unique they think they are, are influenced by everything around them," Cope said. "There's no way a composer can attend a premier of their work and not be affected by a standing ovation or very spotty applause."

Sometimes the influence can be negative, Cope added: Claude Debussy, for example, disliked Richard Wagner's music and produced "a radically new style" through a deliberate attempt to compose along different lines.

The study shows that people "can discern the little things they like about music even in the context of a lot of extraneous sounds," said MIT computational musicologist Michael Scott Cuthbert, who wasn't involved in the research. "But what they don't prove is why music today has changed from the popular music of the past. It doesn't show how changing tastes result in changing music and it doesn't give us a hint of what the future holds for music."

rosie.mestel@latimes.com

Advertisement
Los Angeles Times Articles
|
|
|