Can You Classify What Makes Music Good?
from the objectively-define- dept
About the time I was starting Techdirt, a friend of mine started a company called Mongo Music. I remember getting lunch and hanging out in his living room, which was doubling as his office (surrounded by stacks and stacks of CDs) as he explained the concept. He was going to build up a huge database of music, and use people to define all the various characteristics of the song to identify “that thang” that made music catchy. The idea was that this information was objectively identifiable, and if you liked one song, Mongo’s database would point out other songs that matched it on all the characteristics that were measured. It was an interesting idea – but I wasn’t convinced “that thang” was objectively identifiable. Microsoft disagreed and eventually bought the company – but never really introduced the service Jeremy described to me in his living room that day. Now, it appears that another startup is trying to do the exact same thing. Soundflavor is trying to recommend music based on breaking down each song into more than 700 component parts and having human listeners define each and every one to figure out what makes the song sound the way it does. I’m still not convinced what makes music good can be defined in such a calculated way, but it seems that everyone is trying to figure out the best way to recommend new music to people. Of course, I still stick by what I’ve said in the past. This isn’t a problem that is solved by technology and analysis, but by people. Some of the music I enjoy most was recommended to me by friends who know my tastes and can suggest what they know I’ll like. For example, I’m now completely hooked on a CD from a band I’d never heard of two weeks ago. But, over Thanksgiving I saw an old friend, who knows my musical tastes, and as we were driving somewhere said “Hey, check out this CD I got. I know you’ll like it.” He was right, and the next day I went out and bought the CD myself – and now can’t stop listening to it. It’s the human connection that made this possible. I should also note, by the way, that the CD cost me about $5 and came with not only the CD but a DVD that includes concert footage, interviews and random extras about the band. Whether done on purpose or not, this band gets it. They’re offering you something extra of value for buying the CD.
Comments on “Can You Classify What Makes Music Good?”
No Subject Given
What CD was it?
Re: No Subject Given
Shiner Massive Sound System.
Of course, as was the point of the post, I’m sure not everyone would like this kind of music (which is why I didn’t link to it directly from the post – that wasn’t the point). But if you (as I am) are a big fan of early ska, rocksteady or roots reggae – then it’s a really nice album.
Sampling Error?
When they talk about 700 variables, I’m assuming they classify music according to linear combinations of those 700 variables; a listener would build a preference vector.
The problem I see is that the listener would have to listen to a huge sample size in order to reduce the sample error of the 700 variables to acceptable levels. Additionally, how would it account for multi-modal tastes? What if a listener likes music genres A and B, but not AB hybrids?
Also, how do we form accurate coordinates of the music samples? When they build this database, I’m assuming they have testers populate the linear coordinates. How would the testers agree on consistent score for each variable? It doesn’t sound like it could be automated by machine either, given the subjective nature of music. Are there plans for longitudinal regression?
Re: Sampling Error?
Clever git !!
Interesting real life physical analogy
of a multivariate normal distribution with low covariance values:
http://www.sciam.com/article.cfm?chanID=sa003&articleID=000BD464-9228-1FD7-922883414B7F0000
Are not musical tastes like sand dunes, drifting through the whims of taste as a function of time?
Here we go again
It must be time for people to start reimplementing recommendation systems or something. They didn’t work the last time, why should they work this time? I spent hours playing with the one at MIT (firefly??) and never got suggested anything new that I wanted to listen to. Why should this be any different. And of course as Dorpus points out (even if it was meant ironically (does dorpus do irony??)) ones tastes shift – and you find entirely new genres of music that you find you like. Probably not related to any other stuff you listen to at all. (However human recommendation is not that great either – a lot of people recommend things to me and say “You’ll really like this”, most of the time what they mean is “I like this”, and it turns out I don’t like it.)
Re: Here we go again
> ones tastes shift – and you find entirely new genres of music that you find you like.
fscking-A
I thougth that was the entire point.
Currently, I’m listening to Atari Teenage Riot remixes of Bork that I probably whould have thought to be completely unlistenable at another point in life. …and for once in my life I finally understand the whole genre of dissident music… something my “appreciation of music” class I took in college completely failed to do.
Go figure…
No Subject Given
Didn’t Firefly get bought by Amazon? If so, I’d say it is working just fine. I’ve bought books and CD’s I had never heard of previously because I followed thre Amazon reccomends or toher people are buying links.
Re: No Subject Given
My biggest problem with Amazon’s system is that I already have 95% of what they recommend to me.
all previous posters are fscking morons...
The music industry already has a metric to identify and classify music as “good”: it’s called “Sales”.
Hell, if people are willing to part with money (esp. these days), then the music is probably “good”.
Reminds me of the psychologist who was interviewing a famious bank robber and asked him “why do you rob banks?” in the hope of finding some deep Froidian need that drove them to banks. The answer (blindingly obvious): “Because that’s where the money is.”
I buy music...
…based upon it’s ability to resonate my thorasic cavity.
Do they have a variable for that?
hello