Want to read Slashdot from your mobile device? Point it at m.slashdot.org and keep reading!

 



Forgot your password?
typodupeerror
×
Music Media Businesses

Dissecting Songs Down to Their 'Musical Genome' 368

Carl Bialik from the WSJ writes "The company Pandora Media takes a different tack for its online music-recommendation service. When you tell Pandora a song you like or have bought, it doesn't mine its sales database for records of other purchases by those who have bought the song. Instead, it looks for songs with a similar musical profile, based on a database of 300,000 songs rated on up to 400 characteristics like rhythmic syncopation, vamping and vocal harmonies. To analyze the songs, Pandora has hired Bay Area musicians like San Francisco jazz guitarist Bob Coons. 'When Mr. Coons describes a particular song, he uses phrases like the "complexity of the chromaticism" and "richness of the harmonic structure." He has studied the chord structure in Britney Spears' "Oops I Did It Again," and reports that it is "actually fairly complex," ' the Wall Street Journal Online reports."
This discussion has been archived. No new comments can be posted.

Dissecting Songs Down to Their 'Musical Genome'

Comments Filter:
  • by Rotten ( 8785 ) on Friday October 07, 2005 @02:31PM (#13741435) Journal
    This is one of the signs of the apocalypsis
  • But wati (Score:4, Insightful)

    by Crusader7 ( 916280 ) on Friday October 07, 2005 @02:32PM (#13741441) Journal
    What if I like both eighties hair metal and symphonic orchestra? I guess it's okay to reccomend songs from each of those categories, but as the number of preferences rises, wouldn't it become harder and harder to pick even a specific genre to reccomend, much less a specific album?
    • Re:But wati (Score:4, Informative)

      by Teux ( 737929 ) on Friday October 07, 2005 @02:41PM (#13741536)
      Their system allows you to set different "stations"

      Each station gives you the ability to add a few different types of music, but it's not recommended that you try to mix radically different types. You'll have to use old fashioned judgement to choose a broad category you want to listen to, it does the rest of the work exploring similar music.
    • What if I like both eighties hair metal and symphonic orchestra?

      Aside from the hair metal, thats fine. Just kidding.

      But seriously, I like different styles of music as well, but I don't regularly make a playlist/mix CD that completely crosses all of my music taste. Instead, they are a little more tightly coupled. Maybe that is why this service offers each individual more than one "channel" of music to listen to.
    • If you like metal and symphonies, you might like Trans-Siberian Orchestra [trans-siberian.com].
    • That's easy, you listen to symphonic / warrior metal. Problem solved! \m/
    • What if I like both eighties hair metal and symphonic orchestra? I guess it's okay to reccomend songs from each of those categories, but as the number of preferences rises, wouldn't it become harder and harder to pick even a specific genre to reccomend, much less a specific album?

      The radio stations seem to be based on the search that started them with the "song like this".

      Apparently, you can have multiple radio stations, but not one of mixed genres.

      But, for a new-cool-music identification route, it's pretty

  • by waterlogged ( 210759 ) <crussey@[ ]mail.com ['hot' in gap]> on Friday October 07, 2005 @02:32PM (#13741442)
    ..... Those weren't "Chords" that you were admiring.
  • I typed in "Surfin' USA" by the Beach Boys, and it suggested "Sweet Little 16" by Chuck Berry! It's amazing how similar those sound. Do you think it's a coincidence?

    • Ditto...

      I've been looking for band with a sound like the flaming lips off and on for the last year... it hit on the first try with a song call condition by bleach. sweet.

      • Ditto ditto, although my results were mixed.

        I did a search for a couple of songs, and it gave me some interesting suggestions. "Ch-Check It Out" by the Beastie Boys gave a great track, and "E-Pro" by Beck gave some interesting choices. I got kind of excited and started picking out random tracks in my iTunes collection. Then things went awry.

        I searched for The Charlatans UK (neglecting to put the UK) and it got me in touch with some '60s band called "The Charlatans" that didn't sound right at all. So I added
      • Mercury Rev, The Polyphonic Spree, Brian Wilson's newer stuff, and The Starlight Mints are good places to start. If you like the pre-Soft Bulliten Flaming Lips albums, try Tripping Daisy. There are alot of similar bands in the corridor from Oklahoma City to Dallas.
  • by SailFly ( 560133 ) on Friday October 07, 2005 @02:35PM (#13741474) Homepage
    There is a commercial service that does similar analysis on songs to provide a score based on similar genetic algorithms. As I recall you can upload your own music, and for a nominal fee they will provide the analysis. Apparently many music publishers use this service to find songs from new artists that have a higher propobility of success (wide acceptance). I just don't remember the service, but read about it on-line just a year or two ago...
  • It's early, but I might as well announce it.. flotsm.com: Generates 'tokens' akin to MaxiCode [idautomation.com] based on file content; MP3, PNG, GIF, what have you, colorized/formatted according to file characteristics (file format, bpm, size/resolution, etc) You can match existing 'tokens' for material you like against a database of potentially similar material. The 'token' has a sample of the data stored steganographically, so it's playable, viewable, etc.
  • Wow! (Score:5, Funny)

    by errxn ( 108621 ) on Friday October 07, 2005 @02:39PM (#13741508) Homepage Journal
    I typed in "dog shit" and got Ashlee Simpson's entire catalog! This thing is amazing!
    • I typed in "dog shit" and got Ashlee Simpson's entire catalog! This thing is amazing!

      That's funny. I typed in "tape hiss" and got both her and Milli Vanilli.

  • I'm a musician.. (Score:5, Informative)

    by DJayC ( 595440 ) * on Friday October 07, 2005 @02:39PM (#13741512)
    I'm a musician, and I told it some songs I liked and it's playing a customized radio station of songs that I should like... and it's dead on.

    The best part is that you can ask it "Why are you playing this song", and it will explain it to you.. in terms of the song structure and things like that.

    These are real people analyzing these songs.. this seems like a great service to find new music from bands you don't know. Taking bands out of the context of a "social circle" (like Amazon and itunes do by simply looking at 'people who purchased this also purchased...') is a GREAT idea.

    I urge you to support this project if you are a music lover, or at least check it out and listen for a couple hours.
    • It really works! I tried John Cage 4'33" and didn't get back any results.
    • I gotta say this is a brilliant idea. I've been listening for the last half hour and I'm hooked. I am tempted to sit planted in front of the computer listening for the rest of my ten hours but I can't because now I have to go to the record store....
    • I'm a musician, and I told it some songs I liked and it's playing a customized radio station of songs that I should like... and it's dead on.

      The best part is that you can ask it "Why are you playing this song", and it will explain it to you.. in terms of the song structure and things like that.

      I'm not a musician, but I do listen to pretty obscure electronic music.

      I was shocked that it not only knew the artist I entered (Karsh Kale), but that, as you say, I'm suddenly listening to a custom radio station with

      • heh...I plugged in Juno Reactor, and the third track that came up was Karsh Kale. I've never heard of him, but now he's on the top of my "research further" list...very cool.
    • Re:I'm a musician.. (Score:3, Interesting)

      by anagama ( 611277 )
      I decided to give it a bit of try -- maybe I mixed in too many elements, but my Leonard Cohen/Joanna Newsom/Tori Amos list isn't great -- I don't hate the songs, but then again, only a few I really like. So I was happily skipping along till I got this:

      "Unfortunately, our music licenses force us to limit the number of songs you may skip each hour. Sit back and enjoy the music for now, and click "Guide Us" to let us know what you think of the music we're playing."

      I don't understand that. Why would th

  • I tried this out a month or two ago when a friend recommended it. It actually does a decent job at finding other groups/styles you might like. You basically give it a seed song/group, then it will branch out from there. Based on whether you give the songs it feeds you a thumbs up or thumbs down, it tries to build up a list of common musical traits, which it then uses to feed you more. At any time, you can click a link that will show you why it suggested a certain song (midtempo, mild syncopation, breathy fe
  • by tgd ( 2822 )
    In a couple minutes of playing around with it, I have to say I'm pretty impressed. The first couple groups I put in it didn't know, but when I put Delerium in, it picked out one of the newer Delerium songs I liked and has continued to pick a stream of songs I like.

    I'm impressed. Particularly since the songs it picked are either songs in very different genres or songs I haven't heard of and liked.
  • by Tackhead ( 54550 ) on Friday October 07, 2005 @02:42PM (#13741542)
    > When Mr. Coons describes a particular song, he uses phrases like the "complexity of the chromaticism" and "richness of the harmonic structure." He has studied the chord structure in Britney Spears"

    "Yes, do continue ..." invited the RIAA executive.

    "Oh ... and er ... interesting rhythmic devices too," continued Coons, "which seemed to counterpoint the ... er ... er ..." He floundered.

    Ford leaped to his rescue, hazarding "counterpoint the complexity of the underlying chromaticism of the ... er ..." He floundered too, but Coons was ready again.

    "... humanity of the ..."

    "RIAAnanity," Ford hissed at him.

    "Ah yes, RIAAnanity (sorry) of the singer's publicity-whored-out soul," Coons felt he was on a home stretch now, "which contrives through the richness of the harmonic structure to sublimate this, transcend that, and come to terms with the fundamental dichotomies of the other," (he was reaching a triumphant crescendo ...) "and one is left with a profound and vivid insight into ... into ... er ..." (... which suddenly gave out on him.) Ford leaped in with the coup de grace:

    "Into whatever it was the song was about!"

    The RIAA executive stood up.

    "No, well you're completely wrong," he said, "I just write top 40 music to throw my mean callous heartless exterior into sharp relief. I'm going to throw you off the ship anyway. Guard! Take the prisoners to number three airlock and throw them out!"

    ...

    "...counterpoint the complexity of the underlying chromaticism..." He considered this for a moment, and then unplugged the iVog with a grim smile. "Death's too good for them," he said.

  • O Woops! That was supposed to go to my therapist.
  • Not bad! (Score:3, Insightful)

    by ryanr ( 30917 ) * <ryan@thievco.com> on Friday October 07, 2005 @02:42PM (#13741546) Homepage Journal
    This is the first musical taste-type service I've tried that has gotten anywhere close to accurate. In fact, I've found around 10 of the last 15 or so rather likeable. And they have the Dance Hall Crashers listed, which is a great sign.

    As to questions about "what if you like both foo and bar styles?" You start with one song or band, and it makes a "channel" out of that type. If you want to explore a different genre, I assume you start over.

    It's also full songs, decent quality.

    Overall, pretty nice.
  • I tried this out yesterday. It offered me the same "if you like, you'd like..." stuff as Amazon does. Not really impressed.
  • Artists are constantly being influenced by other artists, how far does this go? Does it go to the point of some song that has similarities to a previous one (according to some database) is not artistic or original? Is it okay to play more than 2, 3, 4 of the same notes in the same sequence as another song without being attacked for infringement?

    Why stop at music? Why not go into visual arts as well? They too have too many similarities that can not be overlooked.

    At this logic, Claude Monet, Pierr
  • by Red Flayer ( 890720 ) on Friday October 07, 2005 @02:48PM (#13741590) Journal
    OK, so they're assigning tags to songs, weighting the importance of those tags, and recommending songs that we may like due to characteristics that may normally be ignored.

    Cool.

    A problem -- there is no way they will be hiring enough professionals to grade every song out there that I might be interested in. If they get a sufficient following, I see labels paying to have their songs indexed... good luck to the independent musicians out there.

    I would hope that they allow people to assign their own weights to different criteria. This is a major problem with most of the automated referral systems. The "people who have bought this also bought X" model doesn't work for me, because my tastes are different from most people... or so I'd like to believe.

    What I'd like to see is a cross-genre analysis of the music that is reviewed. I don't like Pop Country -- so how do I find the Bluegrass I want without weeding through what I consider to be junk?
    • We don't have any concept of "tags" associated with music. We have a room full of musician analysts, and they listen to songs all day scoring each song on ~400 characteristics. Basically, they create a ~400 dimensional vector for each analyzed song. When you create a station based on a particular artist and/or song, you'll hear songs that are the weighted sum of delta squares across these ~400 dimensions. As you give us positive and negative feedback about the music you hear in a station, we bias the weight
  • Only more sphisticated. Several things going on here. It would be wonderful to have the scoring of these songs as played. They you could indeed feed them into Markov models and analyze them. One of the DB admins here did that sort of thing with classical music for his MA years ago with Mozart, Bach, etc. The resultant composer recognizers could correctly identify pieces that were not part of the training sets.

    This sort of analysis might be used in copyright infringment cases as well as looking for new a
  • "He has studied the chord structure in Britney Spears' "Oops I Did It Again," and reports that it is "actually fairly complex,"

    I'm not all that surprised really. I don't care for her music but that's personal taste. While I can't speak for her songwriting ability, I do think she has a talented voice. Sadly, too many people think that simply because they do not like something it must suck and anyone who does like it must be stupid. This phenomenon pervades much more of our culture than just music - art a
    • Studio Musicians (Score:3, Insightful)

      by dmaxwell ( 43234 )
      Pop artists are backed up by a stable of studio musicians and probably songwriters for that matter. Someone like Britney may sing well but she doesn't even have to do that because her voice can be tweaked in real time by equipment. For that matter, the band only has to play somewhat compently although studio musicians tend to know what they're doing technically even if they are discouraged (probably) from applying any of their own imagination. The entire performance can be tweaked in real time now.

      Billy
  • by FirstNoel ( 113932 ) on Friday October 07, 2005 @02:59PM (#13741679) Journal
    I put in "Weird Al", First they gave me "I lost on Jeopardy", then the Kinks with "In a Foreign Land", now it's the Rembrandts - "Just the way it is, Baby".

    While I can see the similarities in syncopation and tone and music feel, it doesn't match the lyrics or the feel of the song. When I'm in a "Weird Al" mood, it's not a Rembrandts mood. The Kinks, maybe.

    Ok, now it's "Tears in the Rain" by Triumph....uh guys...not really...

    It's a novelty. If anything it can give you a jumping off point for finding new bands. It might actually be better served in the "Indie" community. Give them the well known band you like, and it gives you all the related Indie music. That I would like.

    Sean D.
  • by Cally ( 10873 ) on Friday October 07, 2005 @03:03PM (#13741712) Homepage
    There's a rather depressing tendency here on Slashdot for slagging off all contemporary pop (by 'pop' here I mean recordings intended to do little else but lodge in the listener's mind and perhaps convince them to spend money on the artist's recordings.) There has always been disposable crap popular music with little or no merit, it's nothing new. (Read up on the Victorian music hall tradition, certainly in the UK it was overwhelmingly the most popular entertainment for not only the working, but the middle classes too. (Indeed the Gaiety reviews attracted lots of titled Gentlemen and 'crowned heads of Europe'.) For a fascinating work of fiction set in that milleu check out Peter Ackroyd's "Dan Leno and the Limehouse Golem" or Angela Carter's "Nights at the Circus" (which is a little more magical post-realism I suppose, but still excellent.) Some of these cathedrals of mass entertainment are still standing in London, often converted into cinemas or now recycled as modern music venues; the Brixton Academy and the Shepherd's Bush Empire for example.

    Anyway what I was trying to say that there is in fact some GOOD pop out there now and then. If you want to see the talent behind Britney, you need to look at the names of the producers, engineers, song-writers and musicians on the record. The thing I find most disturbing is the, um, let's call it the "racial dimension", especially in the US where music is sickeningly segregated by colour.

    Anyway, miles off-topic, we now return you, etc etc. Sheessh. Does anyone else find Friday evenings profoundly depressing?

  • by merreborn ( 853723 ) on Friday October 07, 2005 @03:04PM (#13741717) Journal
    Vamp: (musical) To perform a cyclical musical sequence, allowing musicians to expand on the basic form.

    At any rate, I'm impressed. I used to use Amazon to find similar music, but that doesn't really work. If you put in an MTV2 metal band, all the "People who bought this also bought..." links are to more MTV2 metal bands. It's hard to break out of the mainstream.

    This, on the other hand, pulled up a bunch of bands that I'm pretty sure don't get commercial *radio play*, much less MTV exposure. Unforetunately, I don't have audio here at work, so I can't speak to the quality of the matches just yet, but I'm sure I'll find something I like that I've never heard of.
  • by tupshin ( 5777 ) <tupshin@tupshin.com> on Friday October 07, 2005 @03:05PM (#13741725) Homepage
    Who the @#$@#$@# put Rick Springfield on my Steely Dan station?

    That's just not right. I think my ears just ran off looking for a new home.

    -Tupshin
  • ...but Britney Spears songs is written and/or arranged by very professionals, and 'Hit Me Babe One More Time' is quite serious work, if we look at chords and how they can be interpreted.
  • by Kainaw ( 676073 ) on Friday October 07, 2005 @03:06PM (#13741737) Homepage Journal
    This reminds me a lot of a research project I worked on years ago... Evolutionary Music and the Zipf-Mandlebrot Law [shaunwagner.com]. Our conclusion back then was that a computer can tell you if music is "pleasant". We didn't want to use 'good' or 'bad' because that would lead to a lot of arguments based on taste, not music.
  • by shoppa ( 464619 ) on Friday October 07, 2005 @03:08PM (#13741753)
    The classic reference is Knuth's The Complexity of Songs [acm.org].

    My favorite part is the end where he references K.C. Sunshine for the song of the least complexity, "That's the way (uh-uh uh-uh) I like it".

  • I think he must have meant the "facial contortions" were actually fairly complex when listening to the Brittney Spears song.
  • Be carefull, the RIAA might try to use the DMCA claiming they are being harmed by this.
  • I've been using Pandora for about a month, and am very pleased with it. It does a great job of picking apparently unrelated artists that fit into the genre of the station you create. The best example I noticed was a station created from Lords of Acid and Chemical Brothers playing a track by Garbage. I don't know Garbage's music well enough to say if all their music fits, but the one track that Pandora picked fit in seemlessly. I listen to it most of the day at work, and maybe an hour or two at night. I've s
  • Here's an idea for combining this model with other models for recommending music. Take the recommend the song based on structure model, combine it with the recommend the songs based on what other people who listen to music like you list to model (e.g. last.fm [last.fm]), and add the Amazon purchasing model to it. Combine that data with specific user feedback... things like what time of day does the person listen to this song, group, genre... what day of the week, what's the weather like (b/c it impacts mood), maybe
  • Common, how much bullshit can you hear/read before identifying it as it. Does it only require jazz in front of the word musician for people to view someone as a musicologist? Britney Spears makes dimwitted easy music for people who don't really like music but wanna look cool. It's what we call in the business train music, it comes real hard and goes fast and then we forget about it, she's no bethoven, she's no Beattles, she's a musician that makes hits and sell records, period. Its quite something to be abl
  • Fractals? (Score:3, Interesting)

    by serutan ( 259622 ) <snoopdoug@RABBIT ... minus herbivore> on Friday October 07, 2005 @03:31PM (#13741975) Homepage
    In the early 80s Scientific American had one of the first articles I ever read about fractal mathematics and music. It talked about a statistical value called Spectral Density, which varies from white noise to "brown" noise. In white noise the signals have no relation from one moment to the next, as in hailstones randomly falling on a piano keyboard. In brown noise they are strongly related, as in a mouse walking up and down the keys. Fractal patterns have a spectral density somewhere in the middle. Neighboring signals stay around each other for a while, then there's a jump to a different area and it stays around there for a while. The jumps themselves show the same pattern. The article said that almost any piece of music that as wide popular appeal, regardless of the genre, has a fractal Spectral Density. Popular pieces of abstract art were also said to have the same property.

    Anyway, I wonder if songs that are similar in the subjective terms Coons uses would be similar in spectral density or some other mathematical way? It would be really interesting to make automated measurements of songs and see if you could get similar clustering.

    Unrelatedly... the article went on to say that the human peripheral nervous system produces white noise, but as you probe closer to the central nervous system the signal becomes more and more fractal, as if the nervous system itself is filtering our raw perceptions and passing a fractal version to our brain. In an experiment with radar scans of a college campus full of people moving around, they found that any one scan was predominantly white noise, but the difference between two scans a second or two apart was fractal noise. They speculated that this might be a key to our ability to process the complex, changing world around us and notice subtle but important details, for example when we immediately notice "something odd" about a person. Fascinating stuff.
  • Spider Robinson's short story Melancholy Elephants [baen.com], in which he discusses the mathematics of unlimited copyright terms.
  • I Bet you they only play RIAA content or song that you can (or Have to) buy.
    What about free music?
  • beta tester (Score:3, Interesting)

    by swinginjohn ( 907770 ) on Friday October 07, 2005 @05:04PM (#13742704) Homepage
    I beta tested this software and it was really fun. A friend of mine is one of the music cataloguers for that company and he got me on the beta. I got some great music recommendations off of it.

    For example, my favorite musician, Ben Folds, had the following characteristics:

    Syncopation

    Singer / songwriter

    Piano lead instrument

    Alternative ... and a few others I can't remember. Then it picks out stuff with the similar characteristics and gives you a little "private radio station" I think is what they called it. Then you can add other artists, songs, albums or genres to give you a little variety. So for example, from my Ben Folds suggestion, I got some selections from Elton John, Joe Jackson, Tori Amos, The Beautiful South, Aimee Mann, etc. etc.

    I then went through and added The Postal Service as a favored artist, then I started getting new flavors added to the mix. Pretty neat.

    The hitch comes from the fact that their recommendations aren't always great. You can skip through their recommendations but you're only allowed to do like 6 per hour. To circumvent this, you can rate each song as it's playing (5-point rating system with the highest being "I really like this sound -- play more like this and the lowest being "Don't play anything like this ever again").

    It's a fun little app. It's nice to just throw on and leave on all day... a good alternative to cheesy shoutcast stations and it's WAAAAY better than the alternative...... corporate FM crap.

    *shudders*

    ~sj

He has not acquired a fortune; the fortune has acquired him. -- Bion

Working...