- News Home
24 April 2014 11:45 am ,
Vol. 344 ,
Major climate data sets have underestimated the rate of global warming in the last 15 years owing largely to poor data...
The tsetse fly is best known as the vector for the trypanosome parasites that cause sleeping sickness and a disease in...
The National Institutes of Health is revising its "two strikes" rule, which allowed researchers only one chance to...
By stabilizing the components of retromers, molecular complexes that act like recycling bins in cells, a recently...
Fossil fuels power modern society by generating heat, but much of that heat is wasted. Semiconductor devices called...
Researchers are gaining insights into what made Supertyphoon Haiyan so powerful and devastating through post-storm...
Millions around the world got a first-hand look at what it was like to be in Tacloban while it was pummeled by...
- 24 April 2014 11:45 am , Vol. 344 , #6182
- About Us
Computer Program 'Evolves' Music From Noise
18 June 2012 5:38 pm
From Mozart to the Beatles, music evolves as listeners get used to sounds they initially find strange or even shocking. As trailblazing music becomes mainstream, artists strike out in new directions. But in a new study, a computer program shows how listeners drive music to evolve in a certain way. Although the resulting strains are hardly Don Giovanni, the finding shows how users' tastes exert their own kind of natural selection, nudging tunes to evolve out of noise.
Bioinformaticist Robert MacCallum of Imperial College London was working with a program called DarwinTunes, which he and his colleagues had developed to study the musical equivalent of evolution in the natural world. The program produces 8-second sequences of randomly generated sounds, or loops, from a database of digital "genes." In a process akin to sexual reproduction, the loops swap bits of code to create offspring. "Genetic" mutations crop up as new material is inserted at random. The "daughter" loops retain some of the pitch, tone quality, and rhythm of their parents, but with their own unique material added.
Previously, DarwinTunes could respond to only one person who would decide which loops went on to replicate. According to MacCallum, that doesn't give a true picture of how music naturally changes—with strange new sound patterns considered satisfying and even becoming commonplace. One explanation, says MacCallum, is that composers and songwriters themselves determine what people listen to, and the audience gets accustomed to it. And with only one person accessing DarwinTunes, this would appear to be the case.
"A single user twiddling knobs to achieve a pleasing sound is like a breeder of purebred dogs selecting for particular traits," MacCallum says. "It doesn't show us how music evolves in the outside world as people listen, pass it on, and recommend it to their friends."
In the new study, appearing online today in the Proceedings of the National Academy of Sciences, MacCallum and colleagues adapted DarwinTunes to be accessed online by almost 7000 participants who rated each sound loop, played in a random order, on a 5-point scale from "can't stand it" to "love it." In a musical take on survival of the fittest, the highest-scored loops went on to pair up with others and replicate. Each resulting generation was rated again for its appeal. After about 2500 generations of sound loops, what started out as a cacophony of noise had evolved into pleasant strains of music.
MacCallum emphasizes that the findings don't dismiss the importance of songwriters and composers in developing innovative, exciting music. "The evolution led to pleasant, jingly tunes that didn't offend anyone but didn't really move anyone, either." He adds that computerized reproduction has its limits. For example, the music did not become more beautiful indefinitely, but hit a plateau at which it stayed at the same, innocuous level.
This plateau effect may be due to the random approach to reproduction used in the original DarwinTunes program, MacCallum explains. In the natural world, the DNA of two parents is not passed on to offspring at random: a child gets one copy of each gene from each parent, which functions when, where, and how they're supposed to. DarwinTunes, which originally swapped musical genes at random, now does so in a more controlled manner and may produce more interesting music, MacCallum says. The music still plateaus eventually, he says, but at a more complex and interesting level.
Composer, musician, and computer programmer David Cope of the University of California, Santa Cruz, is glad to have scientific substantiation of what composers have always known. "There's no way you're not influenced by the fact that your music has influenced others," he says. But he cautions that with a human composer, the influence can work in many ways. He notes that Mozart took audience response personally, but usually continued or even exaggerated musical traits that listeners didn't like. Other composers, including Cope himself, look for new directions when they feel their work is becoming too pleasing.
MacCallum and colleagues are eager to see what kinds of music are ahead for DarwinTunes. Future incarnations, they hope, will enable up to a million users to log on and participate. "Having so many users will help the music evolve a lot faster, and then, who knows?" says MacCallum. Readers can cast their votes at the DarwinTunes Web site.