作譜

I ran across a site called Hit Song Science, which uses the PolyphonicHMI algorithms to determine mathematically whether a song has hit potential. The site is offering four months of free service till Oct. 31, and I didn’t see much in the way of small print to determine what would happen after that grace period expires. I figured I may as well give it a shot with some tracks by Eponymous 4.

However much I may love the songs I create, I have this sense my music is too middle ground to fit any particular audience. It’s too commercial for the indie audience (with whom I have most affinity) but it’s experimental enough to keep it out of reach of mainstream taste. I went into this experiment knowing I probably wouldn’t score very high.

According to the site, a song with a score higher than 7.0 has less resistance to becoming a hit. Scores between 6.75 and 7.0 indicate a borderline hit which requires further marketing push. Anything below that, I guess, you don’t bother.

Here’s how it turned out.

The first song I submitted was "Go", something I wrote back in 1986 but pretty much gutted and rewrote in 2006. I modeled it after Cocco’s "Yawaraka na Kizuato", which sounds like ’90s alt-rock. With a better vocalist and live instruments, I could see this song getting some attention. Hit Song Science score: 6.7.

Hmm. Just 0.5 points shy of a borderline hit. Hit Song Science offers professional reports to critique how a song could be adjusted to become more of a hit, but there’s little indication on the site how much those cost. The FAQ says to e-mail for a quote. That prompt should be made throughout the site.

According to the report, "Go" has an affinity to Rob Thomas’ "Ever the Same". If there’s one thing this exercise makes clear, my hit illiteracy is profound. I don’t recognize any of the titles to which Hit Song Science compared my songs, except "Hey Ya" by Outkast.

Then I tried "What I Deserve", a song I completed back in June. It really, really smells like hit to me, but Hit Song Science isn’t going that far. Score: 6.98.

So far, I’ve agreed with the site’s snap assessments of my work, although it’s a lot more harder on me than I am. And that "Hey Ya" affinity? Yup, this song. Perhaps that’s a strong endorsement?

Then I fed it a bunch of songs I feel would be fan favorites — if I had fans — but not necessarily big hits. First was "Speechless", a piano ballad with very strange percussion effects. Strange enough it would probably never get radio play. And Hit Song Science says … 6.37.

The FAQ says, "We help labels identify those songs that will rise on the charts and sell in the stores with less resistance than other songs." And it also notes, "… a score of 7.30 and higher has solid hit potential. Scores between 4.00 and 7.00 tend to denote moderate to low hit potential."

If I were to bring this song to market, I wouldn’t necessarily position it for chart height, and I would rather maximize sales in areas where it would be most amenable. A score of 6.37 means it’s clearly not a hit for a mass audience, but is it a hit for a highly-targeted one?

Next I tried "Imprint", a song I really like but recognize it has very little hit potential. Hit Song Science handed out its toughest score: 5.99.

I had to agree with this assessment, although a 5.99? That’s a little harsh!

For the next trio of songs, I picked some I could imagine as highly-targeted singles — songs that won’t be popular to a general audience but would be popular to a specific audience. First was "The One to Make Me Whole". My friend chip had actually used an instrumental version of this song as background music for an access television project. I don’t use a very standard chord progression with this song, and that alone could put listeners off. Hit Song Science score: 6.66

Next was "Untold Demons", a rewrite of the very first song I wrote. It’s a bit too akin to an ’80s rock ballad, so I’m not sure how fashionable it is. Hit Song Science agrees: 6.51.

The fact a lot of these songs just scrap under the 6.75 threshold of a borderline hit makes me wonder whether most folks who submit music to the site get this kind of score. The FAQ says, "We use a logarithmic scale where the distance between 4 and 5 is half the distance as that between 5 and 6 and that distance in turn is half that between 6 and 7."

As a result, the "curve" weighs heavily toward a score of 6. I’m guessing other folks who use this service will score mostly with 6.x points — if they’re as mediocre a talent as I am.

The next song really surprised me. It’s a punk song called "Revulsion", and while I think it could be a single for college radio, I don’t think it’s a song for the general public. Hit Song Science is just slightly more ambitious: 6.95

More disturbing are the songs which Hit Song Science thinks has the closest affinity to the song: "I’m Gonna Get You Good" by Shania Twain. I wrote this song under the influence of Number Girl — how can a song which hearkens to a Japanese punk band have some sort of affinity with a country artist? It’s all in the math, I guess.

The next three songs are instrumentals, which usually have no traction on radio. Two of these tracks are actually modern classical pieces not really intended to be hit material. The first is "enigmatics IV", which is kind of a techno-new age thing. Hit Song Science score: 6.18

Not surprising results, but I find myself saying, "Huh? That’s it? C’mon, Hit Song Science, it’s a bit higher than that, right?" I guess I think more highly of myself than I realize.

The absolute shocker is the next track — a six-minute excerpt of Terry Riley’s In C, which I recorded on Ableton Live. Polyphonic HMI claims that their algorithms are based on a wide spectrum of music, from Beethoven to the Beatles. In C has pretty much entered the repertoire, and I wonder if Hit Song Science can figure that out. I think it did, because it scored the highest of anything I submitted: 7.15

Man, I’d love to release that to radio just to see if it really does become the hit HSS says it could be.

Given the high marks it gave a modern classical piece, I tried it with a string quartet I wrote back in high school. I’m no Terry Riley, so I was expecting some sort of 6.x score. Hit Song Science score: 6.38

Nope. Not a hit. But it actually scored better than "enigmatics IV"?

Finally, I tried a cover song of Number Girl’s "DESTRUCTION BABY". I don’t think the original was a chart-making hit in Japan, and the song has never been released in the US. And since my cover tries to stick as close to the original arrangement as possible, perhaps this assessment applies to both cover and original. Hit Song Science score: 6.51.

So does Hit Song Science really tell me anything? I can’t tell.

Most of my scores indicate Eponymous 4 isn’t a hitmaker in terms of a mass audience. Thing is, I would prefer to forge an audience that’s highly targeted. These scores tell you how much resistance a song has to being a hit. Without a full report, I can only guess these "average" scores could very well be good enough for more modest goals.

It was interesting to see what the algorithms thought about my music, and while numbers don’t lie, I can read almost anything into them.