Saturday, January 19, 2013

Their A&R app said "I don't hear a single"

When music lovers don't like a pop song, they often use the word "formula" to express their disdain. If you have artistic integrity, you'll write from the heart instead of painting by numbers, dammit!

Alas. The music world is about to get a whole lot more formulaic. Computer scientists from the University of Bristol have invented an algorithm that can predict whether a song will become a hit.
The researchers used musical features such as, tempo, time signature, song duration and loudness. They also computed more detailed summaries of the songs such as harmonic simplicity, how simple the chord sequence is, and non-harmonicity, how ‘noisy’ the song is.

A ‘hit potential equation’ that scores a song according to its audio features was devised. ... The team found they could classify a song into a ‘hit’ or ‘not hit’ based on its score, with an accuracy rate of 60 per cent as to whether a song will make it to top five, or if it will never reach above position 30 on the UK top 40 singles chart.

So far this is just a bunch of academics who asked an interesting question ("What specific variables give a song mass appeal?") and found an interesting way to answer it. But it has implications beyond the scholarly realm. Potentially troubling implications.

Less innovation

Big record labels will jump on this technology if it works. Why wouldn't they? Historically, they've chosen new acts based on what they think will sell, and what they think will sell sounds like what has sold before. Nothing new about that.

But every now and again, somebody bucks the trend. Bob Dylan and Nirvana weren't putting out what the labels considered commercial music, but they got their shot because an A&R guy sensed that there was something special about them and convinced the higher-ups to give them a chance.

The algorithm can't do that. One of its limitations is that it only makes correct predictions within a given era—a hit from the early '80s isn't the same as a hit today. So if a new sound comes along, something with the potential to catch fire with a large audience, the program has no way of knowing that.

In an ideal world, there could be A&R people who do their own research when a new musical trend sets off their Spidey sense, and then feed their software new data when they're right. Maybe that will happen. Or maybe the big labels, eager for short-term rewards, will just replace their expensive expert humans with software.

I'll leave you to ponder which option is more likely.

Songwriting software

With the ability to predict what songs will be popular, how long will it be until software is actually writing the songs? How long before my fellow songwriters and I are replaced by a machine?

This sounds like the stuff of science fiction, I know. Still, artificial intelligence is getting better and better. And it's not an either/or proposition—there are intermediate steps. Before we have AI that can compose music and lyrics unaided, we may have ...
  • Hit-predicting software as an editing tool. Songwriters will create their own first drafts, then run them through a program to see what they need to change. This isn't the most attractive alternative if you want to express yourself and create something new, but it's highly useful if you're just trying to churn out a hit.
  • Songwriting AI that creates the first draft. Early software won't be able to produce a radio hit on its own. The lyrics will probably sound like a bad Babelfish translation. However, it might hash together something a human songwriter can tweak and build upon.

Wild, handwringing speculation, you say? Well, yes. And it's important to remember that the music industry doesn't control all the music. Thanks to the internet and cheaper recording equipment, it's easier than ever for new bands to gain an audience without the help of a label.

But as far as radio play goes, I really do believe this is something to keep an eye on. Technology gets better, and all sorts of things that once sounded like sci fi have become possible. Someday we may have computer programs writing our greatest hits ... and we won't be able to tell the difference.

Hell, the article I linked to is over a year old.

Maybe it's already happened.


DeppityBob said...

I've seen similar fiction-writing software results. Like you said, it comes across as a bad Babelfish translation. Fiction is more complex than songs, though I can see software writing poetry or flash fiction. Once they get it perfected, you won't know the difference.

MOS said...

I agree with everything you said - and Call Me Maybe is certainly a perfect example. There is just one caveat, tho: It is 60% successful? That's hardly any better than a 50-50 shot.

Brian McDonald said...

Well, they're already working on replacing the singers. I don't see why the songwriter should feel special.

The singer as commodity is pretty much the rule in J-pop, but there are still dozens of amazing bands flying under the radar, then popping up and blowing people away.

cinderkeys said...

Dep: I'd love to see some of those translations. If they have something like that for lyrics already, I'd love to hear the results. They'd be hilarious.

MOS: Depends whether the program's predictions are better than a human's. And even if not, programs can get better with further development.

Brian: True. That's why I mentioned the caveat about indie musicians. I wish it were possible to get internet radio stations on my iPod. I should probably just upgrade my cell phone service ...

cinderkeys said...

Oh, and for the record, I kinda like "Call Me Maybe." But it's an easy target for this sort of thing, yes? :)

cookstone said...

I like your blog. I also think, though it's easy to poke fun at because we heard it everywhere, "call me maybe" is an incredibly well crafted pop tune. That's why it was overplayed, kinda like "Somebody that I used to know" but better. The article above gives a pretty interesting look at the tune & how her album sales may have suffered from the success of the single.

cinderkeys said...

Interesting article. I don't remember "Maybe" getting memed to death, and my Facebook feed doesn't spare me many memes.

Most surprising takeaway: "Kiss is the best pop album of the year, and nobody is listening." Huh. I actually do like "Maybe," even if it's easy to make fun of, but I wonder if I'd want to sit through a whole album of songs like that. Hmm.

Three Chord Monty said...

I have to say I kind of doubt that this software would have been able to predict the hit-making potential in "Harlem Shake."

cinderkeys said...

So the question becomes, would "Harlem Shake" not get made because the software can't predict what people will meme?

Three Chord Monty said...

I think we're going to get MORE things like that--music that becomes popular not because of the music itself but because of some other ingredient--I mean, you could do that gimmicky dance to most dance music. That no software can predict something like that is why people will continue to do things like that, and it's easy enough to do so without regards for what a big recording contract will allow for, either.