The Computer That Identifies The Hits

from the it's-all-in-the-math? dept

A bit over a year ago, we had the story of a company that was trying to break down music into 700 specific components to identify "that thang" that made it a hit. As I explained then, they were only the latest in a long line of companies trying to do so (one of which was eventually bought by Microsoft). However, while most of those companies focused on building a consumer service ("if you liked that song, our algorithms say you'll like this song..."), another company working on similar technology has apparently found a more lucrative business model: selling it as a service to musicians and record labels so they can feel more confident about how well certain songs will catch on. Unlike the company that was going to break songs down into 700 variables, this one only breaks it down into 20 components -- but apparently that's plenty. Record labels and many artists apparently swear by it. Even unsigned musicians are sending in their songs and using high scores from the company to help them get a record deal. While the obvious reaction is that this is just going to make new pop music sound even more alike, the company claims that's not true (though, as you might notice, they'd be a bit biased). They say that the mathematical signatures of songs that are similar may sound completely different. So, a hard rocking band may actually fit into the same classification as a cool jazz singer. While the article discusses the software's successes, it never mentions any failures. It's hard to believe that it's perfectly accurate, so you have to wonder if people are selectively remembering when it was right and downplaying the times it was wrong.