Is there a formula, some special combination of sound codes, that can tell us whether listeners will like a song?
Apparently some people think so. They feel humans are inferior to computers and artificial intelligence in determining what songs you should play.
There’s a tremendous effort going on to build large song databases, coding each tune with such things as tempo, key, and instrumentation.
Perhaps the best known is Pandora’s Music Genome Project (MGP). Musicologists coded Pandora’s 800,000 songs with up to 400 distinct characteristics, everything from chordal patterns to such things as instrumentation and even something called “motion inducing.”
At least Pandora enlisted the help of humans to build their database.
As we highlighted in Want to Build Your Own Pandora? the latest efforts eliminate human input altogether. Computers are replacing Pandora’s musicologists:
A new collaboration between Columbia University researchers and The Echo Nest, a company that tracks online music and delivers listening suggestions to users, hopes to take the human element out of Internet radio. One goal is to deliver better recommendations and more songs through improved artificial intelligence.
Leave it to computer geeks to conclude that the pesky “human element” is hampering song recommendations.
Echo Nest claims to have coded 30 million songs:
Echo Nest has some 5 billion data points for more than 30 million songs in its database, such as artist connections, song similarity, mood, style and acoustic attributes.
You may recall that Clear Channel hired Echo Nest to help create iHeartRadio’s personalizing feature, their version of Pandora.
So is curated radio dead? Will music directors be replaced by computers armed with 5 billion data points to decide what songs to play?
We don’t think so.
Creative music programming is an art, perhaps a dying art, but not a dead art.
The creative part of music programming is finding a balance between the predictable and familiar, with the unpredictable and unfamiliar.
One of the initial complaints with Pandora was that it was too predictable. Every song sounded like the songs that preceded it.
That shouldn’t have come as a surprise. Pandora’s MGP is all about finding songs that match the profile of the seed song.
Of course, the songs are going to sound similar. The very essence of Pandora was to offer up more of the same.
But a listener quickly becomes bored with a string of similarly sounding songs. She needs the right touch of variety. Something that surprises in a pleasant way.
That’s the art that every good music director, every good nightclub DJ understands.
So Pandora responded by adding variety, essentially de-tuning their algorithm.
But it still hasn’t solved the problem. Many Pandora users now just create a new channel every time they sign on in the hopes that the new channel won’t sound as boring as their old channels.
Will computers and artificial intelligence, outperform Pandora’s musicologists and radio’s best music directors?
Will the right algorithm make radio stations sound better?
More likely the opposite. It will make radio stations sound even more predictable and boring.
The trend towards homogenized national playlists has already made local radio stations sound predictable and boring. Layer on top of that an algorithm that makes sure all the songs sound the same.
That will make for some exciting radio.