Slashdot Mirror
Statistically Optimal Music
ShinyPlasticBag writes "'Eigenradio makes its optimal music by analyzing in real time dozens of radio stations at once. When our bank of computers has heard enough music, it will go to work on making more just like it. Since we listen to so much music all the time, Eigenradio is always on and always live. What you hear on Eigenradio is the best of the New Music, distilled and de-correlated. One song on Eigenradio is worth at least twenty songs on old radio.' Listen up here or here (SHOUTcast)."
What does the RIAA have to say about you using their copyrighted material to generate music - music which is arguably not unique, but rather derivatives of their property?
1. Horizontal scrolling required
2. Tiny
3. Virtually no links to anything
4. Very small amount of information
John.
Now, if there was anything worth listening to on the radio, I'd say they'd have something, but hey don't because "Garbage In = Garbage Out".
While hacking up pig snouts and horse hooves might make for an interesting, ummmm... "sausage", it's still nasty dead stuff...
RS
Shoes for Industry. Shoes for the Dead.
IMHO this is yet another example of how academic projects are judged by the amount of attention they attract, rather than on whether they advance the state of the art. This is the reason why people like Kevin Warrick can stick a dog tag in their arm and go around claiming they are the world's first cyborg - all while being lavished with attention by the mainstream media.
All of this leads to an academic system that increasingly rewards self pubicity at the expense of real reasearch.
Oh, BTW - I listened to the radio station, it sounds like a garbled mess - I certainly couldn't determine the point of this from listening to it, but then I could say the same thing about rap.
Interesting that a bank of computers replicating human music can be so much more interesting than humans trying to replicate human music. I guess they have have a long way to go before they can make music as boring as most major record labels. "It's a feature, not a bug."
Maybe partying will help...
And then... hit a college station playing this noise!
What a refreshment! What a way to cleanse the pallette. No chords. No lyrics. No beats. No guitars. Nothing recognizable at all! Just wonderful organized noise.
Then after listening to a LOT of it, especially the stuff that you know was actually composed by a human, something new happens:
You start to listen to the world around you (traffic, nature, conversations) as if it was composed. Imagining a single intention behind the noise of the world. It really is a beautiful mindset. See the restaurant scene in the movie "32 Short Films About Glenn Gould." http://us.imdb.com/title/tt0108328/
If you haven't spent a lot of time with music like this, try it. If you hate it after 5 minutes, listen for 10. If you hate it after 10, listen for 20. Try to appreciate it.
--
Derek Sivers, CD Baby
http://www.cdbaby.com
What makes you think that the slashdot effect changes the content of the music?
I heard the same thing that you did, but as I understand it, the only input that goes into the music is the content of the radio stations to which the server is listening. I don't see how the number of listeners to eigenradio would have cause the effect you're describing.
<sig>Guvf vf abg n frperg zrffntr
Slashdot's first reaction to VMware
What makes you think that the slashdot effect changes the content of the music?
I guess it would depend on how they've configured their systems. If they have one box processing the incoming signals, and another box uploading the result to their Windows Media Server, then we might overload the second box but the sound would be unchanged.
But if the box that does the processing is the same as the one that's attempting to service all the requests from Slashdotters, it seems like it would eat up CPU cycles. That would make it more difficult to do the real-time synthesis of 20 incoming signals. I suspect that's the cause of the toneless drone I was hearing.
Add to that the bandwidth -- do they have one pipe that's receiving 20 signals, outputting (however many) Eigenradio streams, *and* serving up the strangely-formatted web page?
Stressed? Me? Of course not. Stress is what a rubber band feels before it breaks, silly.
Actually this is a serious concept that goes back to John Cage's piece Imaginary Landscape #4, which was scored for 12 radios tuned at random.
Of course that doesn't necessarily preclude it's being satire as well. The mind is fully capable of holding two contradictory ideas at once. Religious fundamentalists do it all the time.
KFG
If you haven't spent a lot of time with music like this, try it. If you hate it after 5 minutes, listen for 10. If you hate it after 10, listen for 20. Try to appreciate it.
Uh, why? I checked eigenradio out a week or two back and, in addition to being boring as hell, it was physically painful to listen to. But I made myself stick with it for a bit, in an attempt to see what was so wonderful about it. I failed miserably.
So, would you care to go beyond your admonition to "try to appreciate it" and tell us just what you think is there to appreciate? Is it just your "beautiful mindset" of believing the world to be overarchingly ordered or is there some other reason you're telling us to continue listening to something we hate?
I think you just defined ambient music.
John Cage
'makes its optimal music by analyzing in real time dozens of radio stations at once. When our bank of computers has heard enough music, it will go to work on making more just like it. Since we listen to so much music all the time, Eigenradio is always on and always live.'
That's all well and good, but what if more than half of those stations happen to be playing music that sucks? (even good stations use filler too..)
I'm slightly scared. This is a technological curiosity of its own might, granted, but this prompts me to envision a rather gloom future. Originally I've thought that the rise of networking would eliminate the entire corporate structure involved in music-making and be replaced with system where everyone can give a go at composing, publish their work online and where the best artists could probably managed to make quite a fortune with voluntary donations.
However, could record companies do the ultimate thing, a la Nineteen Eighty-Four, and create a computer program that produces the music most of us want to hear? Would that mean the end of human creativity on that level of play, or would this algorithm be doomed to failure? It might only take a few years to adjust, and you'd end up liking it.
Of course, a prudent question is, if music can be replicated so easily, what's the point in appreaciating it any longer, as it's clearly something even machines can do well...
Next up: television series writing machines. But, oh wait, we already have reality tv...
Rarely is the need for mulitcast felt more than when a music stream is slashdotted.
Anyone who takes an intermediate signal processing class learns about Princ. Component Analysis (PCA). Loosely, it attempts to represent a set of signals as weighted, linear combinations of sub-signals..... The technique allows you to find the pieces of signal that are common to the overall set. In this case I'm sure they are lining up some radio feeds, performing PCA, doing a little trivial stuff to it, and synthesizing their own "music" based on some transformation of the PCA weights and computed vectors. Not a big deal -- more like a one afternoon project for a grad student, or maybe a class project for a few undergrads...
Computers writing music will never happen. At some level, it will always be people using computers as tools to write music. But we have that already (ie Mixing of music).
:-)) What classifies this as a music composition? It makes a number of algorithmic choices to create a new sound.
First off, this is a single aleotoric (sp?) composition that is extremely similar to John Cage's 'radio symphony' produced a while ago (I don't remember the date or the exact title, but I'm sure someone will correct (or flame) me about it
Even with the lack of posted details about the algorithm, there are a number of assumptions in the algorithm that explain some of the impressions reported on Slashdot.
"Eigenradio plays only the most important frequencies..." - right off the bat, we're assuming that frequencies are important to how we listen to music. Research in psychoacoustics suggests that this isn't the case - we stream music into 'parts' organized by the start and stop points of frequency bands. These streams are then processed for whether the pitch/timbre/rhythm patterns are recognized or not. This is partially demonstrated by the way we talk of 'voices' or 'instruments' having pitch and color (timbre) and of particular songs having 'a good groove'. Any diagram describing this kind of process would have feedback accross the whole diagram, so I doubt its a part of the algorithm used.
"...only the beats with the highest entropy..." Repetition is a feature of all music everywhere - the only musical universal known. Similarly, the 'ideal' degree of entropy in music (how much it repeats) tends to be suprisingly high - music with the highest entropy is actually 'bad'. This differs from culture to culture, but low entropy in good music is the norm, not the exception. Music that has 'high entropy' as a feature already have two strikes against it.
"If you took a bunch of music and asked it, 'Music, what are you, really?' you'd hear Eigenradio singing back at you." This assumes that all music is uniform and can be summaraized into a single source. Contrary to this assumption, there are significant differences between genre types - they exploit different mechanisms for producing pleasure in their listeners. This doesn't even begin to touch non-Western music (even non western pop music). Some of these mechanisms are mutually exclusive (polyphonic music versus homophonic music). An 'average' or 'distilled' reproduction ends up activating no psychological hooks very well and ends up sounding boring.
"They know what you really want to hear. " This assumes that the creator can know what "music" is for you. Each culture hears music differently - with different qualifications for what makes music 'good'. Brain scans of trained classical musicians and their untrained counterparts conducted both in Japan and in the US demonstrated differences in the way these sounds were processed among the four groups. The differences between trained and untrained listeners was radical. Not only do tastes differ, but the music you hear is not the same music I hear - even if the same sound is presented. No single piece of music can legitimately make this claim.
The number you have dialed is imaginary, please rotate your phone 90 degrees and try again.
http://web.media.mit.edu/~bwhitman/10000.html
they seem to be using many from this list, according to the PPS