June 01, 2007

Using Math to Predict Musical Hits

Last week's Science Friday asks Can a song's success be predicted?

Ira Flatow speaks with Mike McCready, co-founder and CEO of Platinum Blue; "McCready says he and his colleagues have come up with an algorithm that will predict whether a song will be a hit." I was very skeptical when I heard the topic of the show, but I was also intrigued.

McCready starts by explaining how record labels decide whether or not they'll release a song as a single: "The first criteria is: does the song sound and feel like a hit? They have A and R professionals - artist and repertoire professionals - people with golden ears who - at the labels who listen to music and decide that.

"The second criteria they use is, are we able to promote the song effectively? That ... covers does the artist have an appropriate appeal to the - audience or the target demographic that they're looking for? ... [Does] the song have somewhere to fit in? Is it written within the zeitgeist of the culture?"

McCready is trying to position his services as a third step -- not to replace the first two steps of human intervention on whether or not a song has the potential to be a hit, but to help them improve the odds. Currently, of every 10 songs they promote, only 1 will be a hit, leaving 9 songs (and $1 million worth of promotion each) in the dust. This is the cool part:

"... [W]e have a computer program that can analyze a fully produced CD and isolate things like melody, harmony, beat, tempo, rhythm, octave, pitch, chord progression, fullness of sound, cadence, sonic variances - about 30 to 35 of these variables that we look at, and we look at how they fit together in the different kinds of patterns that they make up as they come together." McCready claims the company's success rate is 80-100%, including the terrific Gnarls Barkley song "Crazy."

One reason McCready's software isn't more widely used is that, according to McCready, "... [T]he music industry, in a race to adopt new technology, finishes just ahead of the Amish" -- although he adds that there are many music executives who do appreciate his technology.

McCready is also thinking about new markets for his technology, including offering a reduced version of it to musicians to help them create something "sellable", and also "... a music recommendation system in partnership with other technology companies where someone, you know, can go to a retailer and say, '... I like these two songs; what else am I going to like?' "

* Science Friday show, May 25, 2007, Hour Two; mp3 of Ira's interview
* Transcript available on LexisNexis: "The Math Behind Hit Music," National Public Radio Show: Talk of the Nation: Science Friday 3:00 PM EST , May 25, 2007.
* McCready's software is at Music Xray (tm), and his song-seeker software is described at Platinum Song Seeker™.

No comments: