Slashdot Mirror
Decoding the Algorithm for Pop Music
fb4f writes "Over at Modplug, they have an article describing a mathematical algorithm to predict if a given song will become a hit or not. Paraphrasing the article, a Spanish company called Polyphonic HMI has made a business out of analyzing song submissions and predicting their "hitability". Here's their description of the algorithm and here's their FAQ. They claim to have predicted the commercial success of Norah Jones through this method. Here's my question (which is not fully answered in their FAQ): if they (music company executives) are currently using the algorithm to screen submissions for their "hitability", can we (people who listen to music) use the same algorithm to reject recycled tunes and encourage originality? I for one, still like the fresh talent and community feel of the tracking scene."
Someone should make an algo to calculate which Slashdot comments will be moderated up to 5. Should be pretty easy...as long as you bash SCO!
If Singer.Belly.isShown() then mod.singer.+1sexy
If Singer.Voice.isScreaming() then mod.singer.+1punkfav
If Singer.Gender.isMale() then mod.singer.+1prepubescentgirls
If Singer.Label.isRIAA() then mod.singer.+1popular
If Singer.Style.isOriginal() then mod.singer.-1original
Us deciding what is good or the music industry telling us what is good?
This company's stuff doesn't do much good when society is bombarded by what the industry wants us to hear.
It becomes a hit because we don't get much of a choice. ClearChannel plays no variety, the non conclomorate channels don't play variety but instead endlessly repeat that they are not owned by ClearChannel and Infinity...
The only way hits can be decided is through freedom of music.
Support those artists that support the free distribution, copying, and playing of their music. Start your searchs at Sharing the Groove and FuthurNET
Well, I don't know how to tell what songs will be popular. But, obviously, this topic is popular.
40% Action
30% Comedy
30% Romance
0% Madonna
( with credit to Jay Leno )
The fact that no one understands you doesn't mean you're an artist.
They claim to have predicted the commercial success of Norah Jones through this method.
Am I the only one wondering who the hell Norah Jones is?
You damn kids and your pop music. I think I'm going to have to dodder out on the porch and yell at the neighbors' kids for playing on my lawn.
--saint
"Here's my question (which is not fully answered in their FAQ): if they (music company executives) are currently using the algorithm to screen submissions for their "hitability", can we (people who listen to music) use the same algorithm to reject recycled tunes and encourage originality?"
And here's my question: can we use this algorithm to create the hit, instead of determining wether or not it's gonna be a hit?
In need of reliable and affordable server monitoring?
Its strangely insightful. I hate bland manufacturised pop music. It's music for the sheeple. I prefer real music from real bands, like Limp Bizkit and Korn and shit like that. Music for us truely individual people who are outcast from society for daring to be different, for refusing to conform, for sticking a finger up to The Man.
Mapping the higher order functions (if there are any) of the teenage and prepubescent female brain is nigh impossible.
On the other hand, predicting what will be popular might be very easy.
Next big pop hit = whatever the record companies tell them it will be.
Witness the last American Idol. Who did the sheeple choose? The large black guy, Rueben. Months later, who do you hear the most about? The Howdy Doody lookalike who came in second place, Clay.
"you vill like vhat ve vant, not what you vant!"
AFAIK, this isn't new. This technique has been in use for years, at least theoretically.
IIRC, this was first tested on random samplings of classical music. Beethoven and Mozart scored significantly better that others.
Quote : " What do big hits typically score? As stated above we tend to use 7.00 and higher as a score for a hit song because that's where they tend to score. There have been hits that score a little lower but the promotion has tended to be more aggressive. Some big hits score very low on the HSS scale but more than make up for that low score in other aspects of analyses that a label can do on your music with us if you happen to be negotiating a deal. ".
It sound like "we found some correlation, but there is data outside the correlation, and sometimes downright anti-correlation between reality and prediction". I think without looking at the data and the real corelation coeficient between "predicting it will be a hit" and "it was a hit" it is difficult to say anything. And even then, correlation between data does not mean there is causal relation, although *pleasing* to the ear is certainly why we hear at music. I think this kleave other factor out. For example the signification of the lyrics. You ear Mozart uniquely for the pure sound pleasantness, but you do not ear some of the rock/pop for its sound only (try it, many of the greatest hit sound "bland" without their lyric).
Plus even if they try to "reassure" customer in their FAQ, if you comapre things to the past and try to reproduce what has the best functionned in the past, then you will never innovate. Which is IMO the biggest problem now (and it feels that new bands/singer are solely choosen on their look, given prefabricated lyric and tune, and marketed as prima dona, instead of having bands/singer raise on their own by the sheer beauty of their music).
C. Sagan : A demon haunted world:
http://www.amazon.com/gp/product/0345409469/
visit randi.org
So wait a minute.
The algorithm uses the top 30 songs of the last five years as its base of comparison. It then analyzes thousands of songs and determines which ones are most likely to be hits, and those that score best are selectively fed into the market. These songs by necessity become the next set of top 30 hits, and are again used as the algorithm's base of comparison.
So basically, the basis of the system is "these songs will be hits because I say they'll be hits, and I say they'll be hits because they sound like songs that I said would be hits." Isn't this a really, really bad (read: dangerous) case of circular logic?
Anything you might ever need to say about anything has already been said better by Penny Arcade.
It's designed to be a tool for them to protect themselves from mistakes.
...And the endless variations of the Backstreet Boys and New Kids on the Block?
"Well, it bombed in the market, but the software algorithm said it had a chance. I did what the software said was right."
It's your run-of-the-mill corporate bullshit. No creativity, and no real courage to try something different and take a risk.
How do you think we got Milli Vanilli?
The music industry as it is, is little more than a middle-man. Cut them out of the picture, and the consumers benefit, and the REAL artists do too!
Okay, so far, so good; it sounds like they're saying "good music is good music, and here's a tool for telling whether something is good or not." I'm still skeptical at this point, but it's certainly an interesting idea, and one worthy of study.
But then they completely lose me with this one:
IOW, our algorithm says music is good if it sounds like everything else people think is good right now, and if it's different from current Top 40, it's crap.
They make a high-flown reference to the 36 Plots and other serious attempts at artistic analysis, but that's not what they're actually doing. I do believe that good music is good music, good stories are good stories, etc. I can at least consider seriously the hypothesis that all good art has certain qualities in common, and that by analyzing those qualities we can evaluate a new work's chance of lasting success. But the idea that musicians (or writers, or whatever) can keep pumping out stuff exactly like What's Hot Now and be guaranteed a blockbuster is just stupid.
The correlation between ignorance of statistics and using "correlation is not causation" as an argument is close to 1.
The Manual - how to have a number one - the easy way.
Written by the Timelords (the KLF)
(i know, this is a bit offtopic, but hey!, why not?)
A few years ago I saw an interview of a "composer" (forgot his name). They're guys who are somewhat famous because their name is in a lot of albums, but in very small print: they compose the actual music that the band plays. This is standard practice, apparently.
Anyway the point is that the guy pointed out that most pop tunes were rehashes of older pop hits. Maybe you create a different style with different instruments or beat, but the underlying melody is the same. He then showed some examples, in how some modern R&B titles were lifted off some older Rock titles. It's not that hard to believe though, look at how Puff Daddy makes a living out of talking fast over music of old hits.
So in short, one way to predict if a music will be a hit is by creating a database of previous hits and test the correlation...
[and then of course, there's those who say that Classical music tried every combination possible, so nothing can be new afterwards - but that's maybe a little extreme].
The ENIAC Demo Competition
The article states:
"This software will compare the song to a database that contains the "top-30" hit songs of the past five years in order to search for mathematical similarities. The algorithm then assigns each song a score between one and 10. Any song rated more than seven is likely to become a hit."
Now think about this.. use musical eras like the 80's and early 90's as an example because it's reasonably safe to assume this technology didn't exist at that point.
Look at the charts in 5 year chunks, it all sounds the same. In the 80's, everyone either used a synthesizer or had a raging, face-melting solo at some point in the song. Or the early 90's, "grunge" was being pounded into our head incessantly.
It was like that because it was popular. Band X makes it big, and suddenly Bands X1 through X255 appear on the charts mimicing this sound. This seems to happen in, amazingly enough, cycles of 5 years.
Seems to me this software does nothing to show the "hitability" of a song, but rather telling you whether or not it sounds just like what's currently popular, and has been for the past couple years.
Seems about as magical to me as as an algrorythm claiming it can detect boys that like looking at porn.
Al Qaeda has ninjas!
Those of us who were listening to pop music during the 60s enjoyed continuous creativity from many different directions, all going directly into widely popular stuff. Then the music marketing business got its formulas - its algorithms - together in the early 70s and there hasn't been a similar sustained wave of pop culture creativity since. The difference in the 60s, in large part, was that the record companies had seen their old formulas largely stop working about when the 60s began, and so were left to their resourcefulness in finding good stuff beyond their former formulaic sensibilities. By the time the 70s came, a younger generation of music executives had come in who could distill formulas from the prior decade of experience and render rock-based pop largely morbid, as swing-based pop tunes had become by the 50s.
Those of us who live by algorithms should recognize that there are some sorts of human creative intelligence which cannot be captured by formulas, or replaced by them (see physicist Roger Penrose's books on this). If something like this firm's algorithm is really accurate, it should be possible to evolve a neural net to compose pop songs simply by having the success of its efforts defined by feedback from the formula. Would you find living in that world inspiring?
Much of the best of 60s pop music was haunted and quirky. That's what happens when the creative is in the lead, rather than the formulaic. Compare the Elizabethan stage. Human expression triumphs when the formulas, while still there for reference, cease to have a stranglehold over production.
"with their freedom lost all virtue lose" - Milton
So let me get this straight: if a song sounds like a current hit song, it may well be a hit song?
Any this is useful how?
They say they match parameters such as:
- Melody
- Harmony
- Chord progression
- Brilliance
- Fullness of sound
- Beat
- Tempo
- Rhythm
- Octave
- Pitch
This isn't "analysis", it's gross categorization (i.e.:"uptempo pop song in the Michael Bolton vocal style"). It's entirely subjective to the listener - what does "fullness of sound" mean, anyway?Even then, they add this huge disclaimer:
1. The song must be good from an A&R perspective. That is it must sound like a hit song to human ears.
2. It must have optimal mathematical patterns. (that's where this service comes in).
3. It must be promoted well and with an appropriate artist.
Feh. Nothing to see here. If you're interested in real algorithmic analysis, check out David Cope.
"Historically, what is pleasing to the human ear has not changed since man began writing music."
That's true in that music has probably had percussion since the start, other then that, total rubbish.
What about music from other cultures? Totally different scales, numbers of notes, structure, the works. You gonna tell a billion Indian's their taste in music is mathematically wrong?
Music is almost 100% relative. It's about painting a psycho acoustic picture inside the listener. Why do certain sounds feel aggressive, well others are soulful? It's 99% arbitrary.
Goodness, in a pop sense, is a matter of painting a picture a whole bunch of people perceive in a similar way. It's a function of civilization, just like any art.
The very thought that you can mathematically write pop songs. People have been trying for awhile. Even if you get an algorithm for a perfect pop song, everyone would get sick of that style and pop would reinvent itself. It's what happens. Hair metal gives way to Grunge. Grunge gives way to Big Beat, Big Beat gives way to nu metal, nu metal gives way to retro-punk. Hip hop does it all within one genera. Street goes to bling, bling goes to conscience, conscience goes to freestyle street and now we got Outkast doing some sort of 70s funk thing doing triple platinum.
The trick isn't writing songs, that's easy, the trick is writing the songs that work nearly universally.
Ask anybody who does it for a living.
TITLE OF THE SONG - DaVinci's Notebook
Declaration of my feelings for you
Elaboration on those feelings
Description of how long these feelings have existed
Belief that no one else could feel the same as I
Reminiscence of the pleasant times we've shared
And our relationship's perfection
Recounting of the steps that lead to our love's dissillusion
Mostly involving my unfaithfulness and lies
Penitent admission of wrongdoing
Discovery of the depth of my affection
Regret over the lateness of my epiphany
CHORUS:
Title of the song
Naive expression of love
Reluctance to accept that you are gone
Request to turn back time
And rectify my wrongs
Repetition of the title of the song
Enumeration of my various transgressive actions
Of insufficient motivation
Realization that these actions led to your departure
And my resultant lack of sleep and appetite
Renunciation of my past insensitive behavior
Promise of my reformation
Reassurance that you still are foremeost in my thoughts, now,
Need for instructions how to gain your trust again
Request for reconciliation
Listing of the numerouss tasks that I'd perform
Of physical and emotional compensation
CHORUS
Acknowledgement that I acted foolishly
Increasingly desparate pleas for your return
Sorrow for my infidelity
Vain hope that my sins are forgivable
Appeal for one more opportunity
Drop to my knees to elecit crowd response
Prayers to my chosen deity
Modulation and I hold a high note...
CHORUS
As a younger listener, I was well aware of rap music, but it sort of cornered its own market and stayed there. I don't think anyone was prepared for, or could of predicted the massive influx of rap/hip-hop into the mainstream. Personally, it's not my bag of tea (little music is these days). And personally, I don't see what's so prolific about it, other than the fact that a good portion of it has a *very* raw, rebellious overtone that is, for whatever reason, favored by youth. But it's there, it has a huge market, and I find it interesting, if for no other reason than to admire the degree of influence it has had.
Given this, I'm not sure there is any algorithm that can predict what people will decide they like at any given point, as there are so many dynamics at work. As others have pointed out, there is definitely the chance that our music-buying preferences are being manipulated by those at the top telling us what we like. But there are also others - the infamous "what are my friends listening to" I-gotta-be-like-everyone-else mentality. I'd be remiss if I didn't mention another significant consideration, at least with respect to the most of the popular artists: Is there any money in it?
The software thats been cooked basically gives record execs another means of increasing their hit:miss ratio.
So think of it this way, the RIAA claims that they charge high prices to make up for all the flops. They now have a new means to weed out the money wasters. Profit goes up, prices go down... right?
[Fuck Beta]
o0t!