March 29th, 2007

Genometrics in G#

Alan Lomax’s “cantometrics” has long functioned as a Pandora’s box for conversation on the SEM listserv. Yesterday and today, fairly explicitly (literally?). We’ll see what happens tomorrow.

Here’s the current string, in all asynchronous argument –

Alexandre Enkerli
to SEM-L

Mar 27

Fellow music analysts,

To be honest, when a student in my anthropology of music class last year sent us a link to the ” Music Genome Project ” I had something of a knee-jerk reaction and even allowed myself to be a bit snarky. Probably because, to a culturalist, the term “genome” as applied to music seems quite misleading. Since then, I was able to listen to an interview with MGP founder Tim Westegren, which gave some insight into the project’s methodology. More recently, a Radio Open Source show on the effects of randomness in music selections (iPod’s shuffle feature as something of a “breakthrough”) also had Westergren on, talking about his project. Basically, they tag tracks for specific features and allow listener’s to connect those tracks through musical similarities instead of through genre labels. They do have a fairly extensive musical selection, possibly covering some of the musical ground many of us are working in.

To me, the MGP relates fairly directly to Alan Lomax’s work: the analytical method used by MGP resembles Cantometrics and MGP’s Pandora “radio station” seems somewhat related in concept to what I understand Lomax’s Global Jukebox Project to have been, almost twenty years ago. Have some of you looked into the Music Genome Project and its implications for our discipline? Westegren’s work clearly isn’t ethnographic and has no pretense of being ethnomusicological. Yet, we can probably relate to some of the ideas surrounding the project.

Cheers!


Alexandre
http://enkerli.wordpress.com/

Christopher Horgan
to SEM-L

Mar 28

Hello All,

I actually work for Pandora.com. I am one of the Senior Analysts, and I also oversee our Rap and Electronic music areas. If anyone has any questions I’ll be happy to answer tham as candidly as possible.

Here’re a few links with info about Pandora – there’re plenty more interviews around on the web if you do a Google search:

http://pandora.com/corporate/
http://pandora.com/mgp.shtml
http://blog.pandora.com/pandora/

Essentially what we do is to analyse a song based on several hundred musical categories – eveything from vocal register, to the amount of guitar distortion, to the use of sampled music, to the timbre of the drums. We then use the data from these analyses to drive a matching engine that creates playlists of similar music. When using the Pandora radio station you can interact with the matching engine (tell it YES music like this should play on this station, or NO music like this should not play on this station).

It is really an amazing product, and when a user gives the player some direction it is frightening how good the matches are. Also, Pandora is constantly changing which aspects of the song it is matching on so a listener still gets a pleasant amount of variety on their stations.

There are a couple of caveats all should know: Everyone at Pandora works our hardest to ensure there is no bias. We aren’t affiliated with any record label or whatever. We buy and accept QUALITY music of all types – Tho at this point our catalog is limited to: Pop (rock, R&B, blues, punk, etc.), Rap, Latin, Electronic, and Jazz. Classical is just getting it’s feet off the ground and may be going public this year.

Alex – you are correct there are some similarities with Lomax’s work, but I’ll leave it up to the scholars to flesh out this aspect of the discussion further.

Chris Horgan

Thomas Porcello
to SEM-L

Mar 28

Alexandre–

I presented a paper at the 2006 SEM meetings on Pandora.com and the Music Genome Project, and have had stimulating dialogues with colleagues at NYU and Duke about this since. In my view, there is too much smoke-and-mirrors around the coding parameters used by Westergren’s company to make any substantial connection between cantometrics and the so-called musical genome. What are the parameters coded? Which ones seek to the describe the music, and which pertain to the sonic dimensions of sound recording as a medium through which we consume music? How do those parameters grapple with the subjectivity of the listeners who code them? How do the terms of the coding mediate between musical sound and the linguistic conventions of describing music and sound? What does it mean to engage in musical analysis as a means to sell a service to producers and consumers (Pandora) as opposed to engaging in academic analysis of musical structures and performances (Lomax)? Can one compare a marketing tool to disciplinary analysis? Can one compare a method geared toward delivering taste-based consumption to a method for (ostensibly) describing relationships among musical structure, culture, and society?

In any event, I think such comparisons miss the point of the Music Genome Project. There is a common appeal to scientific authority in both Westergren’s genomes metaphor (which is little more than a metaphor, if you look at the biological definition of genome) and Lomax’s work (regardless of what position you take with respect to it), but little more. Ultimately, what is of interest about the Music Genome Project are not any of its specific claims, but the way in which it participates in a debate about whether marketing music to consumers should be based on (ostensibly) scientific authority, or on populist definitions of what constitutes good music (as it exists in social networking sites like last.fm).

Tom Porcello

Associate Professor, Anthropology
and
Director, Media Studies
Director, Independent Program
Vassar College, Poughkeepsie, NY
**

2 Comments Add your own

  • 1. Birdseed  |  March 31st, 2007 at 2:14 pm

    I’m not a musicologist at all, just a hobbyist music historian and listener, but I thought I’d share my opinion on Pandora anyway. :)

    The problem with the system is that it treats music merely as sound rather than a bearer of meaning, cultural significance and identity. There’s no way to limit by subject matter, era, relations to other music or “undergroundiness level”. For example, when listening to hip-hop I much prefer the dance-oriented fluff to the “meaningful” (I was about to say “conciousness”) underground sounds, and the two are very different geographically, socially, textually. But Pandora wouldn’t necessarily differentiate between them. Lucky they’re sufficiently good at doing hip-hop for them to be able to differentiate between the different kinds anyway… My southern hip-hop station is fairly good by now.

    That brings up a second problem though. It’s very unbalanced genre-wise. It’s great for some stuff (I’ve got a marvellous jazzy soundtrack station and a good instrumental surf-rock station), while being abysmal for other stuff. (Have you tried making a reggae station?) The content just isn’t complete enough, and that goes for the desription categories too – they’re great for a genre like hip-hop (which is why the results are good for me) but awful for genres outside their matrix of consideration.

    I’ve spent ages trying to make a decent eighties electronic channel for instance. The description parameteres currently in Pandora are not focussed enough to tell the difference between Hi-NRG and nineties house (or whatever), and the content for that kind of genre is really bad compared to, say, nineties indie or sixties jazz.

    DESPITE all this it’s clearly the best system on the net for this sort of thing, mainly because it recognises that no-one wants to listen to the same music all the time…

  • 2. Birdseed  |  April 1st, 2007 at 2:49 pm

    Typical example of the frustration Pandora will sometimes put you through:

    Create new station based on artist “Miquel Brown”. Tracks that appear: Nineties dance track (downvote), Miquel Brown track (upvote), nineties dance track (down), Hazell Dean track (up), recent house track (down), Trinere track (up), nineties dance track (down), “I’m In Love” by Evelyn Champagne King (yes, yes, yes, finally getting somewhere, up!), nineties dance track (nooooooo! down, down, down), trance track (can’t skip tracks any more, gives up).

    All of which could be avoided by including definitions like “primitive digital synthesizer sounds” (yes), “significant use of dynamic filters” (no) or “housey basslines” (no). Or, even better, timeframe definitions!

Leave a Comment

Required

Required, hidden

Some HTML allowed:
<a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <strike> <strong>

Trackback this post  |  Subscribe to the comments via RSS Feed

Wayne&Wax

I'm a techno-musicologist, internet annotator, imagined community organizer.

I left my <3 in the digital global, but I reside in Cambridge, MA, where I'm from.

I represent like that.

wayne at wayneandwax dot com

Tag Cloud

academic aesthetics af-am africa anthro arab art baby babylonia beatresearch blogging bookish boston brazil cambridge caribbean chicago commerce copywrong cumbia dance dubstep ethno ethno europe events funkcarioca gigs global globalghettotech hip-hop humor industry internet interview jamaica jazz juke kwaito latin lifey linkthink mashup media mexico middleeast mixx nation newyork panama politricks pop public puertorico r&b race radio reggae reggaeton remix riddimmeth0d rock sampling seasonal sexuality soundscape tech techno traxx tv UK video whirledmusic worldmusic youth

Month

 

Creative Commons License

chacarron chacarronchaca-riggity-ron