Pandora’s jukebox

“Can you help me discover more music that I’ll like?”

This is the driving question and prime directive of a fairly new web-based music service called Pandora, which offers streaming music in a “radio station” format. The listener can create up to 100 different personalized “stations” based on an artist or a particular song. To start, you type in an artist name or song title, then the player searches Pandora’s database and streams songs they have determined are similar to what you have chosen. As Pandora plays, you can give each song a “thumbs up” or “thumbs down,” which will then also help the player select music more appropriate to what you want to hear. The database purports to hold some 300,000 songs from 10,000 artists.

The service is free, but banner ads have already started up. If you can’t stand them, an ad-free subscription is $36 a year. Record companies do not pay to have their properties included in the database (at least according to the FAQ), so just cool your payola paranoia. We live in consumer-oriented times, and you can click straight through to Amazon or iTunes and still purchase an album or song.

Being a music snob, I initially felt compelled to test Pandora before allowing it to help me “discover” more music. I set about trying to stump it by putting in the names of somewhat obscure artists. Without getting too incredibly obscure, I only stymied it a couple times (and it wasn’t with Silversun Pickups, in case you’re wondering).

With that hurdle cleared, it was time to see what I could learn from Pandora. My first extended experiment was with Tom Waits, whom I consider a cornerstone to my musical tastes.

I started by trying to steer the player, to give each song my approval or disapproval so that it would refine its selections to the point at which it would only play songs I like. The measure I used was, “Would I turn the station if this song came on the radio?” Every time I said “yes,” I’d give a thumbs down, and every time I said “no,” I’d give a thumbs up. Using this rule I found myself giving the thumbs up to songs that I didn’t particularly like, but to which I didn’t particularly object either.

After about an hour, I’d painted myself into a corner. Other than Waits, the player stuck to the same five artists over and over, with only an occasional variation. These were José Gonzalez, The Microphones, The Mountain Goats, Richard Thompson, and Elliott Smith.

I like all of these artists and was impressed that Pandora deduced my interest in them from my liking Tom Waits, but I’m hard-pressed to come up with a category that fits them all (other than that they’re all solo male performers with their own distinctive styles). To the uninterested or uninformed, they may all sound the same, but to me there are obvious differences.

I felt like I’d hit a dead end, so I tried another artist and a different tack. I decided just to let the player run without any guidance on my part and almost haphazardly picked Pink Floyd as my next artist. While I like them and own a few of their records, they’re not a monument in my musical development. Some of their stuff I like; some I don’t. I’d just seen Akron/Family, though, and they’d struck me as sounding similar to early Floyd, so they were on my mind.

Pandora started with “Scarecrow” from Piper at the Gates of Dawn, Pink Floyd’s first record and only release with their legendary founding singer and eventual mental-health casualty Syd Barrett. I’d expected something from The Wall or even one of the post-Roger Waters (read hopelessly un-cool) records. For a music snob like me, starting with a Syd Barrett song was about as cool as things could get. I was impressed and got that little lift I get from having my tastes validated.

After a few interesting but not necessarily surprising selections — mostly psychedelia — the player edged (and then plunged) into different territory. Elton John’s “Rocket Man” was followed by Jackson Browne’s “Your Bright Baby Blues,” and then . . . and then, well, then eventually “Diamond Ring” by Bon Jovi.

Bon Jovi.

I was disturbed. Elton John is fine. I like Elton John in only a slightly ironic way. Jackson Browne, well Jackson Browne can be listenable, kinda sorta maybe, in that mellow ’70s rock sort of way. But Bon Jovi? There’s just not enough irony in the world for me to like a Bon Jovi song.

Not only does Bon Jovi seem pretty damned removed from Pink Floyd (any era, but especially Syd Barrett era), but I am not even a potential fan of Bon Jovi (“I’m not I’m not I’m not!!!”) and I resent anyone or any computer database that even implies it.

About this time I started to pay closer attention to a key feature on Pandora, the “Why is this song playing?” link, in the “Guide Us” menu. I clicked on it in a “what-the-hell-were-you-thinking” frame of mind.

This is what it said:

Based on what you’ve told us so far, we’re playing this track because it features electric rock instrumentation, a subtle use of vocal harmony, mixed minor & major key tonality, acoustic rhythm guitars and many other similarities identified in the music genome project.

Okay, check, check, check, all of those elements are there in “Diamond Ring” and certainly sound like elements that are in “Wish You Were Here,” another Pink Floyd song that the player had dealt me earlier. But still, like the kids say, WTF?

And then I finally awoke to the underlying logic of Pandora and the people behind it, The Music Genome Project. The Project (an exposé might call it a “shadowy organization”) is run by a group of musicians and “technologists” who came together in Oakland in 2000 to “capture the essence of music at the most fundamental level.” Curious to discover what elements of songs listeners respond to, they decided to identify and catalogue the attributes or “genes” of music.

Taken together these genes capture the unique and magical musical identity of a song — everything from melody, harmony and rhythm, to instrumentation, orchestration, arrangement, lyrics, and of course the rich world of singing and vocal harmony. It’s not about what a band looks like, or what genre they supposedly belong to, or about who buys their records — it’s about what each individual song sounds like.

This is not at all a service like Amazon’s “customers who bought products by this artist also bought x.” The MGP’s members listen to each song they put in the database as objectively as possible, focusing on identifying as many musical elements as they can and categorizing each song accordingly, leaving what’s not in the song out of their calculations.

Many objections immediately sprang to my graduate-school indoctrinated mind.

While isolating music from its cultural context in order to classify it by “musical attributes” may be an interesting exercise, it’s impossible. What counts as a musical attribute, or even what counts as music, is culturally conditioned and limited by what we’re exposed to. And why try to isolate music from its history and culture, really? Music is culture, after all.

From my usual perspective, isolating musical elements like “a mix of acoustic and electric instrumentation” and “extensive vamping” seems hokey, simplistic, even sterile. I’ve never thought, “Gee, I like this song because of its major key tonality and meandering melodic phrasing.” It was initially difficult for me to see how this kind of classification could enhance my listening pleasure.

But I plowed on and persevered. Free uninterrupted listening, the possibility of finding new music, and my wounded pride drove me on. And eventually, setting aside my objections as best as possible, the Music Genome Project and my Pandora experiments forced me to think about my musical tastes in a different way.

Generally, I value free-form radio and mixed CDs for the eclectic and unexpected that ends up sounding really good: “Whoa! Tuvan throat singing segues into Patsy Cline in the coolest way! Who knew?” I’ve always snobbishly thought that there’s no real way to map and connect the elements of all the music I like — jazz of all sorts, blues, country, avant-garde, classical, punk, zydeco, indie, alternative (or whatever you’d call it), etc.

I’ve got to admit, though, that music can be broken down into sound elements. “Diamond Ring” doesn’t really sound like “Wish You Were Here,” but it does share some identifiable traits.

Who knows? If they keep working on it, maybe by the year 2525 the MGP will identify the ineluctable modality of the audible that somehow really does connect Tuvan throat singing to Patsy Cline. That would be something to see and hear.

Returning to my reaction to the Bon Jovi song, even though my hackles had been raised (not really ever to be lowered) I had to question my own tastes. Why is it that I like “Wish You Were Here” but not “Diamond Ring”?

Is there an identifiable musical element that rules one in and the other out? Or is it just cultural bias developed from my desire to please all the other music snobs I’ve known, admired, and tried to emulate?

Frankly, I’m not entirely sure.

Perhaps, then, Pandora has offered me a really valuable lesson that I didn’t expect. I started by wanting to learn about more new music (and I have — I must admit that Pandora has taught me about many artists in the last few months), but I learned more about listening for the specifics of what I like and dislike.

Every now and then I’ll have to focus and listen for that special something that I adore or detest. Is it in the song itself? Or in what I think about the band and its fans?

I’m still not completely comfortable with this way of listening, but I kind of like the unsettling feeling.

I hope it lasts.

33 responses to “Pandora’s jukebox”

  1. Rachel says:

    When I first read your article’s heading, I thought it was about our contributor Pandora’s jukebox–which would be eclectic and fascinating indeed.

    Your point about music being inextricable from its context is well-taken; cracking (and then changing up) the culture code is what a lot of hipsters seem to live for. How else could the Strokes be cool in 2001 but not 2002 (or, God forbid, 2006!)? How else could someone love the Buzzcocks but hate Franz Ferdinand and the Arctic Monkeys? Why is it OK to ironically appropriate the populism of, say, Kelly Clarkson (“Since U Been Gone” totally rocks) but not Bon Jovi?

    And how do some bands pull off perennial coolness?

  2. Scott Godfrey says:

    Mr. Wager, great post; oh how you give love a bad name!

    Anyway, The Music Genome Project reminds me of the Komar and Melamid project in which they surveyed 1000 Americans about the attributes they would most like to see in a painting, and painted it. K&M’s project, and I think your experience, illustrates that art cannot be fully appreciated or understood by being broken into its component parts – go shopping in a junkyard and you get junk, or Jovi.

  3. hey, tim. welcome aboard.

    i figured out pretty quickly that i couldn’t use pandora to generate background music for my work day: its selections turn out to be too uniform. i’ve enjoyed it in ways similar to what you describe — as a game, to satisfy my curiosity about what it will choose next. but like you, i prefer freeform radio precisely because of its unexpected congruities, and that’s not something i’ve found pandora capable of. (pandora the jukebox, not my friend and TGW contributor PB.)

    it *does* get used around our house,. though. anna (12) uses it every day. if you open it on any of our computers you’ll autostart with kelly clarkson or avril lavigne.

    i should add that i once downloaded “wanted dead or alive” because i thought it might work on a mixtape. nope. just couldn’t do it. i think someone could make a really cool cover of that song, though.

  4. Stephanie Wells says:

    Actually, I have way too much to say about Pandora and its project to write at this second, so will add more later, but for starters let me just say that when I told Pandora my favorite song was “Moonage Daydream,” it gave me “Bungle in the Jungle.” I KNOW!!!! What an outrage!

    So at first I thought this perfectly illustrated your point, but then I was forced to admit as I listened that they actually have similarities, which was even more unsettling than the initial match. It caused me to say, disgruntledly, “Harrumph.”

  5. Scott Godfrey says:

    Steph, I laughed so hard that “snotts are running down my nose.”

  6. brooke says:

    I wish I had more time to comment on this post. My Master’s thesis involved researching these music recommenders and building something somewhat similar. While I am usually impressed with Pandora’s recommendations (that is, what they play based on your listening and ratings), there are several problems with their approach.

    The biggest problem is scalability. They are relying on a team of musicologists to encode every song with metadata. Thus, after 6 years of coding, they only have 300,000 songs in their DB. Even if that number was ten times larger, it still only represents a small fraction of the musical universe. I would venture to say that the number of songs released per year far exceeds their capacity to encode it wiith the relevant metadata.

    This problem could be temporary. Pandora uses human knowledge to classify songs, but there is a great deal of research underway to teach machines to classify music. That is, given a particular audio signal, use machine learning to extract features such as pitch, tempo, melody, key and so forth, and use that to group songs into categories. One can imagine that such technology, if fully developed and used in the context of human oversight, might speed up the classification process.

    But that brings us to the question of authority. Why would I trust someone (or something) to classify a song? Particularly someone who may not think like me, or an algorithm I didn’t write? What I call rock someone else might call country, etc. and so forth.

    Until recently, Pandora didn’t have a very useful ‘feedback’ loop from the listeners to the playlist generation. They fixed that recently, which means that if you disagree with a particular classification, you can vote your opinion on it. This is a major improvement.

    But I’m puzzled as to why they have not elected to open up their ‘genome’ database to their users (actually, I’m not surprised since it is a proprietary system, which came out of the Savage Beast project). This closed system misses the opportunity to use its user base to improve the metadata. After all, many of us ‘music snobs’ know a ridiculous amount about the music we listen to. Why not allow use to tag our music they way we want, and incorporate the resulting folksonomy into the playlist generation process? Oddly, folksonomies tend to produce richer metadata, for reasons I won’t discuss here.

    Actually, such a service already exists, called Last.FM. I happen to love last.fm, and use it frequently to find new music. These guys went the other way, opting for more of a collaborative filtering approach to recommendations. CF has its own set of problems, often skewing recommendations away from the long tail, towards more popular music. Perhaps a combination of these two services will one day hit that sweet spot. One such service, a mash-up of pandora and last.fm, was already created a month or two ago. It got shoddy reviews and I haven’t used it. But maybe some variation on that idea will be the next big thing in music recommendation.

  7. PB says:

    Reading my name about a million times. Cool.

  8. Tim Wager says:

    Thanks, all, for your comments. Pandora (Brewer), when I decided to write this post I thought you might be a wee bit confused by the title, etc. Glad to know that you’ve enjoyed seeing your name over and over.

    Brooke, I’d heard of last.fm, but haven’t tried it out yet. I will, though. Partially, I hope that Pandora *doesn’t* refine the classification problems you cite, though of course inaccuracy isn’t the aim of projects like this one. (Ramping up the scale would be great, though.) I have come to *enjoy* the “mistakes” it makes like playing Whitney Houston on the Massive Attack station. It forces me to question, if only briefly, how these artists might sound similar.

    Recently I’ve taken to trying artists I’m “supposed” to like on Pandora. I’ve read and heard many times, for instance, that I should really like Scott Walker. I’ve tried, believe me. Now that I’ve heard him on Pandora in the context of Neil Diamond, Antony & the Johnsons, Blue Nile, and other artists I *do* like (to varying degrees and for different reasons), I have a newfound appreciation for him.

    It’s true, Bryan, that uniformity can get to be a problem. On my Walker “station”, for instance, I’ve heard the same songs again and again (and I haven’t listened that much), and the player never really strays from playing mid-tempo orchestral pop, even though it’s not supposed to classify by genre.

    I like to think of Pandora as offering *a* way to listen to music, a supplement to my other listening habits, not a replacement for them.

    And Scott, what? You gonna let me get away with taking a crack at the Jersey boys?

  9. have you been listening to the new scott walker album? i’ve played it a couple times this wk. it’s a rewarding listen — weird, but rewarding. i’m digging around for some older ones — the only other thing of his i had was the one song on the life aquatic OST.

  10. Brett Bobley says:

    I really enjoyed that article. I liked the point you raised when you said:

    Why is it that I like “Wish You Were Here” but not “Diamond Ring”?

    Is there an identifiable musical element that rules one in and the other out? Or is it just cultural bias developed from my desire to please all the other music snobs I’ve known, admired, and tried to emulate?

    Frankly, I’m not entirely sure.

    This is a very interesting notion and I’m not sure either. The cultural bias certainly does play a major role. I find it interesting to study this issue across decades. The Beatles where an immensely popular band. Their initial fanbase was largely young girls, certainly not music hipsters. Not entirely unlike the kids who, today, might like Britney or maybe a teen-pop boy band or whoever is in the charts today. And yet, despite their popularity, today’s hipsters can say with confidence that the Beatles were one of the greatest bands ever (along with many other mega-popular bands of the times like the Who, Stones, etc). By contrast, I can’t imagine being a fan of an equally popular band today. When was the last time you bought an album that was in the top ten? Was popular music just “better” in the 50’s and 60’s? Is today’s top ten stuff just utter crap? Or am I just being a biased dumb-ass faux hipster? (In the 60’s I only listened to the Velvets and the Plastic People of the Universe, man!).

    Perhaps it has something to do with the development of a real music underground. Today there is an infrastructure in place to support underground bands that simply wasn’t there in the 60’s. Maybe there were other bands in Liverpool that kicked the Beatles’ asses but because there were no indie labels/fanzines/ipods/pitchfork/myspace etc, they simply couldn’t survive or get noticed.

    Shit, that Kelly Clarkson song does kinda rock…. If only she wasn’t on American Idol then I could groove to it without feeling guilty.

  11. sure the beatles started as a bubblegum boy band, but it didn’t take them long to demonstrate that they were artists capable of completely transforming the genre they worked in. i can’t think of any pre-fab pop musician today who can make the same claim. the most artistically aggressive “pop” music i can think of from recent years is outkast — and even then, kids tend to like their singles and ignore the rest of their albums.

  12. Tim Wager says:

    I haven’t heard anything from the new Scott Walker. I don’t have any of his older stuff, either, so I can’t really advise. If you plug him into Pandora, you’ll hear plenty from “Tilt” (from 1997) and a selection from the box set, which compiles (though to almost no one’s satisfaction) the 60’s and 70’s stuff. (Just got Einsturzende Neubauten on the Scott Walker station. Hmmm. Music snob sez, “thumbs up”.)

  13. he sounds like he may have been antony’s father. antony’s mother, of course, would have been bryan ferry.

  14. brooke says:

    Tim,

    I agree, one of the nice things about recommenders is the ‘surprises’ one gets. Call them mistakes, anamolies, or whatever, it is fun to come across unexpected relationships in the recommendations. If we could anticipate all of the recommedations ahead of time, it wouldn’t be so fun to play with these services. They are fun precisely because we learn about new music or unknown relationships between songs or artists.

    I don’t think the goal should be to remove these at all, rather to establish a system that can find the best possible surprises. One example of this might be called ‘genre-jumping,’ (a term I just made up), which would involve making suggestions that cross most common conceptions of similarity, based on some more obscure metric. In our project (called Orpheus (this site is out of date, unfortunately…)), we mined several sources of information and allowed a user to explore artist relationships via several paths. One path was artist collaboration. This is a good example of ‘genre jumping’ because it permitted a user to see that artist A collaborated with Artist B on Song C. This is, for example, how I discovered my love for Cat Power. She played on a Handome Boy Modeling School album (rock/hip-hop). When I saw the link, I could then go and explore the Cat Power node, which introduced me to all sorts of artists that I hadn’t thought about.

    The point is, I think services like Pandora are heading in the right direction, but their goal should be to collect as much metadata as possible, and allow the user to decide which metrics are important to them, and how extreme or obscure they want their recommendations to be. One other service that I forgot about, but that does illustrate the value of machine learning, folksonomies and listening habits in generating recommendations is Music Strands. Perhaps they are a bit esoteric, but interesting nonetheless.

    Thanks, by the way, for a great post and for generating such entertaining comments!

  15. Rita R says:

    “Is there an identifiable musical element that rules one in and the other out?”

    As in some formal element, e.g. a chord arrangement? It might be interesting to think of comparisons with software used for textual identification, the kind of program that was run on Primary Colors and various early modern sonnets (what is the ‘identifiable textual element’ of Shakespeare and is it found in this text?)

  16. AA says:

    I started the article by thinking “who the hell is tim wagner and why does he have the same sensibility as me” to thinking seriously about the music genome project.

    As a post-graduate school victim as well I ask, how on earth can we separate music from context? I admire the aim of the project, but it would not help me explain why I absolutely hated Guns ‘N Roses when they were alive, but I sort of like them now because of their headgear–Axle Rose’s ridiculous headscarf and Slash’s stupid top hat. Or why I was a huge Clash fan in the day but now find their self-righteousness tiresome (it ain’t Coca Cola, it’s rice….)

    Or, indeed, how can we drink Coca Cola without the red and white swoosh. Or eat McDonalds without “lovin’ it”?

  17. Tim Wager says:

    I started reading your comment by thinking “who the hell is AA and why is he/she misspelling my name” to thinking seriously about how attempting to separate music from context might help us focus on its technical aspects as objectively as possible before allowing other factors to influence our responses to it.

    Fashion and politics are of course elements of music culture and of course influence the way we think about bands. But these factors are also external to the sounds that actually constitute the songs. Your responses to Guns ‘N Roses’ headgear and to The Clash’s politics are subjective and ‘non-musical’. This doesn’t make them invalid in a discussion of bands you like and why, but it does demonstrate how there are multiple factors that have nothing to do with music itself that influence our responses.

    In discussions of what songs are better and why, one hears many cop outs such as “well, that’s your view and this is mine and everything’s subjective.” Sure, everything has elements of the subjective to it, but without something that we can agree on as objective, how can we talk to each other about anything? Pandora tries to create and/or identify objective statements about music, which to my mind contributes to a better way to discuss music.

    I’ll cut this short now and go out on a limb here to say that Pandora and other classification projects like it in their own small way also contribute (by extension) to bettering social and political discourse by attempting to foster an objective way to discuss cultural production. There, I said it.

  18. AA says:

    Again, I wonder, who is tim wagner, and how is it that he shares my sensibilities?

    If the word “objective” is to have any meaning, and as a word it must (right Dave B?), its application should be open to discussion. Ignore those who say musical tastes are “subjective”.

    I think your blog touches on the broader theme of music’s role in our own identity, or more precisely, as our own identity as consumers. I was certainly one of those in college who abandoned U2 at Joshua Tree because they “sold out”, flagrantly borrowing the Rolling Stones’ Exile on Main Street strategy of mining American roots music. I was proud of that stance at the time, and used it to sort potential girlfriends. I can divide my friends into two groups: those who fell upon Radiohead’s OK Computer and loved it; and those who read the hype, listened to Radiohead’s OK Computer, and didn’t really care for it.

    To the point, and the reason why I liked your blog. We consume music in the same way we consume soft drinks. As an advanced culture, that means we consume through identification. The music genome project is in some ways like the blind cola taste test, but we do not pay 47 cents per can above the cola production cost because of how the cola tastes. We pay that because…well, that to me is the Big Question.

  19. now that *i* know who the hell AA is (and am wondering if this moniker feels safer because no food product graphics can possibly be attached, though the Al Anon reference is intriguing) i have to remind you that i fall into the category of friends who first heard OK Computer at your house, sitting round the pool one summer, and said out loud, “Shit. That could have been the best U2 album since Joshua Tree.”

  20. Tim Wager says:

    Well, AA, seeing as how there’s still no Tim Wagner here, I guess I’ll just have to write his comment for him.

    Hype is such an amazingly tainting component of music appreciation. I am particularly resistant to it but often wonder if I’m just doing a reverse-twist, i.e., not listening objectively (a highly charged word, of course, and one open to debate) because I’m perversely contrary when it comes to what the herd is doing. I like, strike that, LOVE not liking what “everybody” likes.

    _In re_ Radiohead, well, here’s what hype can do to me. When “Creep” came out and was splashed all over everywhere I thought, “Cool murky guitar sound right there, but otherwise, well, standard pop rock that’s fairly interesting but nothing special.” I ignored or did not have thrust in my face anything else off “Pablo Honey” or “The Bends.” Mercifully, I was out of the country when “OK Computer” hit big. When I got back all of my friends told me that I had to had to had to listen to this record. I did, but couldn’t really hear anything through the hype, quite possibly because I’d thrown up my “don’t-believe-the-hype” screen and didn’t want to — or even couldn’t possibly — ‘hear’ the record for all the noise surrounding it. When “Kid A” came out I just went out and bought that sucker for some reason. And loved it. Then I went back and listened to the other records again and liked “OKC” fine (having lowered my hype deflectors). It’s a good (not great) record, but previously I couldn’t hear the good record in there because everybody was screaming at me to hear the great record. Furthermore, in my foray into their back catalogue, I learned what everybody here should already know, that “The Bends” is actually their best record. End of story.

    Lesson? Disliking a record because there’s a bunch of hype about it is just as impure a relationship to it as liking it because there’s a bunch of hype about it. (Not that I’ve ever really taken this lesson to heart or forced myself to change my mind about some things. I stand by my response to all Wilco after “Being There” — soporific and pretentious.)

  21. AA says:

    Hype, or criticism more generally, is like a 15-year-old son experimenting with drugs—you might want to kick him out of the house for his erratic, self-destructive behavior, but you can’t because you still love him.

    So how to handle him? You can either tune in and become hype’s victim, either by following the hype or adopting the hype-deflector approach. Or you can tune out. But if you do, how do you search out what’s good and new? Asking more informed friends doesn’t work, because like you said, they’ve all got their hype-deflector shields up. Do you employ an anti-hype deflector shield and search out music your friends all say sucks? How do you walk the fine line, which is what I think you’re trying to do?

    Bringing it back to the genome project, we are social beings, and we like or don’t like music for cultural reasons beyond its genomes. Otherwise we would all be listening to Bon Jovi. “objectivity” cannot be divorced from our diverse cultural experiences, which is what makes the word so much fun when properly employed.

    (oh, and how can anyone argue with a straight face that “The Bend’s” was Radiohead’s best album?)

  22. Tim Wager says:

    I like your analogy and feel the pain of the dilemma.

    And well, here I am with a straight face saying that “The Bends” is Radiohead’s best album. I’ll not bore our other participants with my well-reasoned, rock-solid argument, but will gladly make it to you off site, preferably in person should I ever have the pleasure to meet you, AA.

    T

  23. Alex Ross at The New Yorker just put up an entry on his blog about Pandora’s difficulties with classical composers. You can find it here.

  24. brooke says:

    Extreme Technology just published a useful comparison of several music recommendation services, including Pandora, Last.FM, MusicStrands, LivePlasma, and about 4 others. Check it out right here

  25. Lil'snake says:

    Hate to suggest something practical in the context of such an interesting conversation. But hey it’s been a few days since the last comment about cultural context.

    Here are couple of things that made my first few tries at Pandora stations more listenable and interesting:
    (1) Define stations using specific songs rather than artists, include a certain amount of musical breadth, and let the cultural/generational breadth happen via musical cross-references. How? Start a station around a focal song/arrangement/performance. Then broaden it out with additional recordings that complement the initial song. Example: I put in several tracks that sample styles of composition and performance on Sarah McLachlan’s first four albums, added a couple traditional and original numbers performed by Gillian Welch, same for Emmylou Harris, some Bjork, included “Didn’t leave Nobody but the Baby” (Harris, Alison Krauss, and Welch), and a 10,000 Maniacs cut. Made a very listenable stream, with variety, found performances scattered across 3 decades, plenty of artists I want to check out, had some predictable similarities (cuts by Enya and Natalie Merchant turned up within a couple of hours), generated some surprises (Prince’s Tangerine?), and established a sort of artistic spectrum for the playlist algorithms. This was much more satisfying than the relatively dull sequence of piano ballads that linked starting from a Sarah McLachlan song that Pandora picked as a prototype.

    (2) As Tom Wager pointed out, it’s kind of easy to paint yourself in a corner. Make only very selective list of ‘thumbs down’. If a particular song and anything musically similar really don’t belong in the musical context, then click “I don’t like this song”. If something is mildly annoying or just blah, skip ahead instead. So far on my Sarah channel, I’ve vetoed only two selections: a honky-tonk number by Sammy Kershaw and Yanni’s “Aria”.

    For another, rather different, ‘station’ I started with White Light / White Heat (Lou Reed’s live version) and added a variety of recordings spanning the transition from hard/acid/glam rock through punk and grunge. Right now the similarity space seems a little sparse. I can hear the gaps. Still tweaking that one.

    For a visual metaphor of such a similarity space (showing artisits rather compositions/performances), see liveplasma.com.

    – Recent convert

  26. Tim Wager says:

    There’s a piece in today’s NYT Arts & Leisure section about Pandora and the effect that it and other web-based services are having on music and music marketing. Check it out: http://www.nytimes.com/2006/09/03/arts/music/03leed.html?_r=1&ref=music&oref=slogin.

  27. bryan says:

    congrats on your wedding, tim. now, admit it — are you reopening this thread in an attempt to work your way up to 62?

  28. Tim Wager says:

    Am I that transparent? I must confess that I rejoiced that the NYT provided me the opportunity to pump up the comments on my post. And thanks, by the way, for taking the bait and helping me get up to 28. Only 34 to go!

    Thanks, too, for your congrats. It was a grand time and a lovely honeymoon, too.

  29. Scott Godfrey says:

    Welcome home Tim, and as a wedding gift of sorts, here’s number 29.

  30. Lisa Tremain says:

    Yay. Happy marriage.

    Good luck with the comments competition. Are we picking teams here?

  31. Tim Wager says:

    At the risk of being accused of pumping up my own comments numbers (though this post is nowhere near many others and it seems to me that we’re well beyond record keeping anyway), here’s yet another story about Pandora-like music recommenders, from Slate:

    http://www.slate.com/id/2149977/.

  32. […] Best of Tim Wager: “Pandora’s jukebox” […]

  33. Tim Wager says:

    Many of you may already know about the legislative threat to net radio services like Pandora. Some of you may not. Here’s a link to a piece on Pitchfork that will give you the lowdown and link you to savenetradio.org, which has a petition going.