Queen Comes To Predictive Software
Radio listeners freaked out when Queen’s six-minute magnum opus Bohemian Rhapsody debuted in the 1970s. Now the sprawling rock anthem is proving too much to handle for a new group of listeners.
Listening programs that try to classify songs based on their sound patterns have a tough time with Bohemian Rhapsody’s storm of disparate musical sections, but an electrical engineer is finding a way to get around this little problem.
UC San Diego Ph.D. student Luke Barrington confronted the problem while creating his project to create an application that classifies music not by song or band, but by what instruments are played in it, what genre it comes from or what emotions it inspires. The purpose of this is to be a different kind of Pandora; instead of asking a program to provide new songs based on artists you already like, you ask it to provide new songs that fit an instrument and a mood.
As you might imagine, it’s difficult for software to tell whether a song is romantic, joyous or rebellious, so Barrington’s solution was to build up a database for comparison.
As you might imagine, it’s difficult for software to tell whether a song is romantic, joyous or rebellious, so Barrington’s solution was to build up a database for comparison.
His Facebook app Herd It plays songs for users and asks them what instruments they hear and how the song makes them feel. The computer can then connect the physical properties of a song’s sound to its emotional impact, as reported by legions of rocking Facebookers.
That’s all well and good for a Ramones song, which the computer could probably identify as two minutes of greatness. But things are a little trickier for Queen.
So Barrington programmed Herd It to break heterogeneous pieces of music into their component parts. Thus, the machine classifies Bohemian Rhapsody’s gentle piano opening, it’s Galileo! Galileo! operatic midsection and its raucous guitar solo separately, making it easier for the computer to understand just what this sonic conglomeration is and how it can be recommended to users.
Now that we’ve gotten to the point that our mechanical friends, and not just our human ones, can recommend new tunes, it’s nice to see that more complex music won’t be left out. It remains to be seen, however, whether a computer’s definition of “romantic” matches our own, but since it draws its opinions by combining all of ours, it will probably be fairly accurate.
However, the danger I fear is that combining the likes and dislikes of a lot of people will result in the same thing movie or music producers create when they try to appeal to everyone: mediocrity.
