Software Predicts Music Success

Frankenbuffer writes "The Globe and Mail today reports that MIT researchers have developed a computer program to analyze pop music and predict how people will react to it. The method, developed at MIT's Media Laboratory, analyzes the pitch, rhythm, and other characteristics of music. What makes the technology unusual is that it also takes into account social responses to hit music gathered from weblogs, chat rooms, music reviews, and other online discussions, and correlates this data to the music to guage the popularity of a particular sound. According to the researchers, the software has accurately predicted Billboard hits for the past several months."

Comments

Indie Artists

[punkdigerati]

Can Independant artists who want to see how well their songs are going to do be able to use this software for themselves?

Re:Indie Artists

[bad-badtz-maru]

I'm not being pedantic but I do want to mention that independant labels have some fantastic pop artists. Obviously, the artists aren't "popular" by mainstream sales figures, but they share a sound similar to mainstream pop music only the accompaninment is typically more avant-garde and the lyrics a bit more educated. "Bis" would be the first group that comes to mind. Freeform Five is another top-notch indie pop band, many cuts off their album "Strangest Things" would sound right at home on mainstream radio. The excellent quality of indie pop, and the fact that it will never see mainstream airplay in the US, is really the constant reminder for me of how bad US radio has become. These artists do get played on Sirius and XM, however. I have been very surprised at how progressive the playlists of some of the satellite radio stations are.

More targetted version

[kebes]

I'm not sure I care whether or not this software can predict the next pop craze... but I wouldn't mind a more personalized version. Maybe something that analyzes my mp3 collection, and then automatically checks new releases and looks for tracks that correlate well with my preferences? It would be nice to have a system that pulls out the things I'm most interested in. And a piece of software would be more impartial than the media executives and promoters who want to sell me the "next big sound."

I think lots of people would love something like that. If iTunes automatically integrated technology like that, I think it would be a hit. And best of all, it would level the playing field. Small bands could "get noticed" by the common person if their sound was something that the given person liked.

Now If

[miyako]

Now if they could just crack the algorithm the music industry uses to generate music, all of the geeks of the world could create free music that sounds just like Britny Spears/P Dilldy dooldy whatever/Pop or Rap Group D'jour....
No wait, on second thought I'd rather keep my sanity.
Joking aside, this sort of research might be interesting from a psychological point of view. If they've developed an algorith that can tell what music is "good" it seems like with some proper research it might provide some insite into the way the brain process music, which could help scientists to better understand the way the brain interprets patterns, etc. If such an algorithm could be used to generate "good" music, it might be useful for things like games, where the game could provide parameters based on what's going on, and algorithmically generate appropriate music.

Pop tarts

[w.timmeh]

Great. So now the recording labels will be able to take a demo, run it through the software, and know exactly how bad a contract they can afford to offer you. Yet another reason to publish independently.

Fortunately, this does seem to only apply to fluffy pop music. Bands in less market-driven genres will hopefully continue to rely on good song writing and support from their audiences to drive sales.

Re:great...

[rust42]

A (quite successful)songwriter in Australia was recently interviewed and said he had worked out the basics of writing a hit,e.g. so many bars intro, chorus of such and such type etc. and had proved it worked. Now his aim was to write hits and popular songs that fell outside that specification.

Killing art? Not really.

[dex.pdx]

Even though a system such as this would seem to have the logical result of killing the art of music, it simply can not.

As a musician I have noticed that the more I study music theory (theory is really just the language to describe music) the more I come to realize that almost all music is the same. You can spot similarities between four hundred year old classics and current "Indie"/Pop/Rock/Whatever, the connection is there.

So what might you ask makes it different?

The style makes it different. The way the individual artist performs/arranges/records a piece. Not to mention that lyrical content adds a whole new dimension?

Saying something like this will ruin the art of music is like saying that the grammar checker in any document editor ruins the art of writing. Though it is neat that a program could possible sort out "popular" music, which just means it's able to emulate the human ear just a little bit.

Greg Egan wrote a story somewhat along these lines

[joe user jr]

Excerpt:

When they started making music straight from the Azciak Polls, everybody howled about the Death of Art -- as if the process was anything new, anything more than an efficient closure of what had been happening for years. Groups were already assembled on the basis of elaborate market research. The Azciak Probes were already revealing people's tastes in breakfast cereals, politicians, and rock stars. Why not scan the brains of the populace, discover precisely what music they'd be willing to pay for, and then manufacture it -- all in a single, streamlined process, with no human intervention required? From the probes buried in a random sample of twenty thousand representative skulls, to the construction of the virtual bands (down to mock biographies, and all the right birthmarks and tattoos), to the synthesis of photorealist computer-animated videos, accessible for a suitable fee ... the music industry had finally achieved its long-cherished goal: cutting out everyone but the middleman.

Ok, a little less prosaic than the item under discussion, but an interesting read...

Read the whole story, at: http://www.infinityplus.co.uk/stories/worth.htm

If you're interested, Greg Egan's site: http://gregegan.customer.netspace.net.au/

Not Impressed

[David_Shultz]

I know it's generally more interesting to talk about how cool it is that technology is doing such great things, and to start imagining all the implications of the wonderful progress we're making. I'm going to do something dangerously uncool and uninteresting -I'm going to expose this article for what it really is. And what is that? An overhyped project with nothing impressive going for it. Really. Nothing. Let's examine the claims to fame made in the article.

First of all, the big claim made is that "the system has been predicting Billboard hits with surprising accuracy over the past several months." They achieve this one simply by reading weblogs. That is ridiculously simple to do: count the number of times a song is mentioned. Predict based on this value. They have not claimed that they are able to predict success on NEW songs based on scanning the music, they have simply claimed, as above, that they can predict billboard hits. Another easy method for doing this is as follows: Take todays hit-list. Copy it identically. That's tomorrows hit-list. There is not that much change over a day -you will get pretty good accuracy.

Second, the software picks out little known groups with popular sounds that agents have missed (ex. Crossfade). How does the network achieve this amazing (sarcasm) feat? By noticing a spike in focus-group data; "HitPredictor struck gold again in late 2003, when its computers flagged a blip in the focus-group data." Hoorah! MIT has invented an algorithm that finds spikes in data.

Thirdly, the software describes the music it hears, from "sexy to romantic to loud and upbeat". Again, this is not complicated. Just look at weblogs and attach to any song titles nearby descriptor words from a small set. This of course is just an example of a simple way to do it. I'm sure there are others.

But there must be some reason for us reading about it? Mustn't there? Well yes. Firstly, there is profit to be made by individuals offering services. Secondly, you can't write or post stories titled "MIT project achieves nothing interesting".