$Hit Song Science
A few weeks ago, something called "Hit Song Science" by Polyphonic HMI (Human Media Interface) was all over the news. The service claims to use an Artificial Intelligence to analyze songs with an algorithm and compare them to a database they've compiled of hit songs since 1956 to determine whether or not they have "hit potential."
I believe that AIs have come a long way in recent years, but I also believe that a company will use the term “AI” instead of “program” because it sounds like it can’t be bought, replicated, or (most importantly) understood. Audio processing has definitely become extremely powerful too, but something about the reporting on it didn’t settle well with me. In addition to the all the spelling errors on their webpage, Polyphonic HMI doesn't say if their analysis has ever been proven or independently confirmed. They often refer to "optimal mathematical patterns" that is common to most hit songs, though in the various news reports and in the FAQ on their webpage they only cite one example of a success they predicted - Norah Jones. I’m a bit of a stickler, so I wouldn’t mind seeing proof that they actually predicted her success, as opposed to simply testing her music later and finding out it fit in the scheme.
They also allow themselves plenty of outs, saying that this service is only a tool that must be used in conjunction with human ears, and the song must be promoted well with an appropriate artist. OK, problem number one: “the appropriate artist.” So are they examining the written music, and not the texture that an actual recording provides? How many versions of “I Only Have Eyes for You” have I heard? Because the only one I ever liked is by the Flamingos. I won’t even get into something as basic as tempo, and that’s something that will change quite a bit with every performer.
But they seem to be testing the actual recorded material, because written music does not encompass all of this:
Maybe there’s more to their submission process, but many of the articles and the company webpage say that people and studios submit CDs of the music. Obviously there’s a lot of audio technology that is way beyond my understanding, but to my knowledge, there aren't algorithms that can process a song and determine even half of attributes they suggest.
How about a testimonial? How about one person that swears by it? Surely someone in the music industry believes it really helped. This leads to another one of their outs: "the song must be promoted well." One thing I've learned working in television is that promotion can be spun many, many different ways - usually to cover up a failure. Everyone in entertainment is in search of their audience, and it's very easy blame bad or inappropriate promotion. Sometimes it's true. Sometimes promotion can make someone useless a hit. It's a pretty big and hugely vague variable.
I want to hear some negative responses from them too. I want to hear about someone big who had a huge failure just as they predicted. Have any of their clients been told that they weren’t hit songwriters?
Finally there’s the price tag: over five grand per CD submitted. It’s just low enough that extremely desperate-for-success musicians can afford it, but certainly an acceptable wage for an artificial intelligence (somebody’s got to feed the little apps).
Unlike much of outrage this service has incited, I’m not another whining musician (in fact, my talents as an actual instrument player are pretty sad). Nor do I think there’s some kind of magic in music that can never be quantified. I like science, I like automation, and I even like the idea of some music being made entirely by an AI.
I also have a pretty good nose for bullshit.
I believe that AIs have come a long way in recent years, but I also believe that a company will use the term “AI” instead of “program” because it sounds like it can’t be bought, replicated, or (most importantly) understood. Audio processing has definitely become extremely powerful too, but something about the reporting on it didn’t settle well with me. In addition to the all the spelling errors on their webpage, Polyphonic HMI doesn't say if their analysis has ever been proven or independently confirmed. They often refer to "optimal mathematical patterns" that is common to most hit songs, though in the various news reports and in the FAQ on their webpage they only cite one example of a success they predicted - Norah Jones. I’m a bit of a stickler, so I wouldn’t mind seeing proof that they actually predicted her success, as opposed to simply testing her music later and finding out it fit in the scheme.
They also allow themselves plenty of outs, saying that this service is only a tool that must be used in conjunction with human ears, and the song must be promoted well with an appropriate artist. OK, problem number one: “the appropriate artist.” So are they examining the written music, and not the texture that an actual recording provides? How many versions of “I Only Have Eyes for You” have I heard? Because the only one I ever liked is by the Flamingos. I won’t even get into something as basic as tempo, and that’s something that will change quite a bit with every performer.
But they seem to be testing the actual recorded material, because written music does not encompass all of this:
"Some of these event [sic] are patterns in melody, harmony, chord progression, brilliance, fullness of sound, beat, tempo, rhythm, octave, and pitch."
Maybe there’s more to their submission process, but many of the articles and the company webpage say that people and studios submit CDs of the music. Obviously there’s a lot of audio technology that is way beyond my understanding, but to my knowledge, there aren't algorithms that can process a song and determine even half of attributes they suggest.
How about a testimonial? How about one person that swears by it? Surely someone in the music industry believes it really helped. This leads to another one of their outs: "the song must be promoted well." One thing I've learned working in television is that promotion can be spun many, many different ways - usually to cover up a failure. Everyone in entertainment is in search of their audience, and it's very easy blame bad or inappropriate promotion. Sometimes it's true. Sometimes promotion can make someone useless a hit. It's a pretty big and hugely vague variable.
I want to hear some negative responses from them too. I want to hear about someone big who had a huge failure just as they predicted. Have any of their clients been told that they weren’t hit songwriters?
Finally there’s the price tag: over five grand per CD submitted. It’s just low enough that extremely desperate-for-success musicians can afford it, but certainly an acceptable wage for an artificial intelligence (somebody’s got to feed the little apps).
Unlike much of outrage this service has incited, I’m not another whining musician (in fact, my talents as an actual instrument player are pretty sad). Nor do I think there’s some kind of magic in music that can never be quantified. I like science, I like automation, and I even like the idea of some music being made entirely by an AI.
I also have a pretty good nose for bullshit.
Hear my music or find out more at
0 Comments:
Post a Comment
<< Home