The Most Wanted Song

I’ve been thinking about the emerging economy surrounding “musical intelligence” software. Through the systematic comparison of a new song to a database containing millions of songs from the past, this software makes a prediction about whether or not the new track will be successful. The pioneer in this field was the Spanish company Polyphonic HMI, who market their service as “Hit Song Science.” Some executives at Polyphonic HMI left to start their own company, Platinum Blue Music Intelligence, which promotes its service under the name Music Xray. Platinum Blue’s software visually plots an individual record as a point of light in a three-dimensional “music universe,” with each point placed in proximity to other records with similar sonic traits. Once the data had been visualized in this manner, the people at Platinum Blue observed that the vast majority of chart-topping hit songs were clumped together in approximately fifty “clusters.” The closer a new song lies to one of these clusters, the more likely that it will be a hit. Musical intelligence software is used by record companies hoping to minimize risk, by amateur musicians hoping to convince labels that they have what it takes for mainstream success, and by online radio providers like Pandora, whose “Music Genome Project” recommends new tracks to users on the basis of an algorithm comprised of 400 musical components. Music intelligence software now shapes creative choices in the recording studio and helps to determine which tracks a record label will promote, and so can easily be perceived as the latest evidence of the utter desperation and moral bankruptcy of the record industry. Some assert however, that the software can convince reluctant record executives to get behind new and innovative artists, provided that their “HSS” scores are high enough (the oft-cited example here being the claim that software predicted the success of Norah Jones’ CD Come Away With Me (2002) in the face of industry skepticism).

The widespread adoption of services such as Hit Song Science, Music Xray and the Music Genome Project suggests that scholars interested in popular culture need to pay more attention to emerging articulations between cultural analytics and the cultural industries. In fact, the discourse surrounding Pandora frames the Music Genome Project as a response to Cultural Studies. A writer for the New York Times notes that in contrast to prevalent “social” theories of musical preference, Pandora’s data-driven approach “ignores the crowd”: “The idea is to figure out what you like, not what a market might like. More interesting, the idea is that the taste of your cool friends, your peers, the traditional music critics, big-label talent scouts and the latest influential music blog are all equally irrelevant. That’s all cultural information, not musical information. And theoretically at least, Pandora’s approach distances music-liking from the cultural information that generally attaches to it.” What Pandora’s system ignores, the author concludes with a shrug in the direction of Pierre Bourdieu, is the social dimension of taste. How might scholars interested not in filtering out cultural information, but in bringing it into sharper focus deploy this same software? Popular musicians have long been known for “misusing” new technology, for throwing away the instruction manual and deploying the latest gadgets in unforeseen ways – using digital samplers for example, not simply to capture a more faithful cello sound, but to cut and paste old Parliament records. Can humanities researchers make a similar move and hot-wire the latest hit song science to create “musical knowledge” in addition to “musical intelligence”? In addition to the continued interest in “social” theories of musical culture, scholars might have something to learn from digital analyses of sound form. Press reports describe how software tends to uncover surprising similarities between artists commonly considered to be vastly different – U2 and Beethoven for example, or Van Halen and MOR piano singer Vanessa Carlton. Could hit clusters in the “music universe” reveal new genres, or new histories of popular culture? What insights might be gained if we expanded the sonic data set beyond the Billboard charts, to include types of recording besides popular music, or tracks made outside of the United States and Western Europe? What significant clusters might appear in an analysis of the recordings found in the University of California, Santa Barbara’s Cylinder Project, which contains over six thousand digitized sound recordings from the first decades of recorded sound?

Those interested in the different models of “hit song science” might look for inspiration to Dave Soldier’s thought-provoking 1997 People’s Choice Music project. Working with Komar and Melamid, Soldier composed two songs determined by a survey on musical preferences. The least popular musical traits as determined by the poll were synthesized into the 25-minute “Most Unwanted Song,” which features wild variations in tempo and volume, an operatic soprano rapping, plugs for Wal-mart, and a healthy serving of bagpipes. More interesting is “The Most Wanted Song,” whose lite-rock synths, meandering saxophone, and romance narrative is both hilarious and eerily familiar. Think of it as the sonic equivalent of Platinum Blue’s digital visualization of a certain well-trod corner of the musical universe.