January 2, 2006

Pandora.

The Music Genome Project. I love it!
Posted by Ann Althouse at 7:46 PM

7 comments:

Anonymous said...

As a lover of classical music and jazz, I found the pickings non-existent in classical and mightly slim in jazz. I want to hear Coltrane, not something "like" Coltrane.

This service also shares a fault of iTunes: Not all music can be categorized as "songs." I think the creators of this project need somewhat broader musical horizons.

1/2/06, 8:51 PM
Alex Nunez said...

Theo, there is room for improvement, but on a whole, I would have to rate Pandora as being pretty darn great. I've been using it for a while now, and have been surprised at how well it matches rock/blues songs with my tastes.

1/2/06, 9:08 PM
Meade said...

Why is this song playing?

Based on what you've told us so far, we're playing this track because it features folk influences, mild rhythmic syncopation, mixed acoustic and electric instrumentation, a vocal-centric aesthetic and major key tonality.

Based on what you've told us so far, we're playing this track because it features bop influences, a piano solo, an acoustic bass solo, light drumming and a brisk swing feel.

Based on what you've told us so far, we're playing this track because it features a subtle use of vocal harmony, mild rhythmic syncopation, repetitive melodic phrasing, extensive vamping and many other similarities identified in the music genome project.

What a blast!

1/2/06, 9:14 PM
Joe Hogan said...

I've just tried Pandora on Ann's recomendation and it seems amazing. I entered Steely Dan as the starting point and, having asked only for the deletion of one song, the first 8 - 10 songs are all on my mp3 player. Assuming they didn't read my song files, they have a pretty astounding logarithm. The only hesitation is that they included 3 Steely Dan tunes in this first batch.

1/2/06, 9:39 PM
chuck b. said...

This seems to work better for me when I feed it the names of artists that are less well known than obvious big names. Kathleen Edwards, New Pornographers and Devendra Banhart all worked better than Lucinda Williams, Radiohead, and Nick Drake. The former set gave me back a lot of stuff I’d never heard of, whereas the latter referred me to stuff I already knew.

(Although the Rolling Stones lead me all over the place; Bob Mould, The Left Banke, Harry Nilsson... it also started off w/ a relatively obscure Stones tune)

I tried John Coltrane too. I would not expect to find anything as good as Coltrane that I don’t already know; there just isn’t as much commensurately good jazz as there is commensurately good rock. But when I entered Django Reinhardt and Cecil Taylor, I got back more interesting, unfamiliar stuff.

And when I got to the part where they asked for money, I deleted cookies and resumed where I left off w/out trouble. When I started skipping a lot of songs (I knew everything that came back w/ Dinah Washington), it explained their license limits the amount of music I can skip per hour. Interesting.

1/2/06, 10:21 PM
Troy said...

Thanks Ann, that is a great site. Who knew that New Order and solo Robert Plant sounded similar? I have both albums and never really made the connection (at least between the 2 songs Pandora put on my New Order station.

I have a new timewaster! and iTunes scout.

1/2/06, 11:11 PM
sierra said...

I've used both Pandora & Last.fm, and both have their good points. Pandora's particularly good at getting around the musical cliques that manifest in social network services. For an example, here's Last.fm's list of artists similar to Caetano Veloso, consisting overwhelmingly of Brazilian acts. If you happen to like an Alt-Rock act that's musically similar, you won't make the connection across social domains. And I doubt if Last.fm would have connected New Order and Robert Plant.

On the other hand, some of Pandora's output is puzzling to the point of distraction. It tends to favor rarer criteria, so if you tell it you like Bob Dylan it's more likely to recommend songs with harmonicas in them than songs with unusual lyrics. In running that test, I was surprised to find there were no female acts with a "prominent harmonica part." To get around that bias, I specified a Dylan song ("Every Grain of Sand") that was also covered by Emmylou Harris. That resulted in a flood of irrelevant mainstream Country & Western. So that tells me they're not just considering qualities of the song common to both performances; they're doing an OR, not an AND.

Some of the output's just plain weird. An Emmylou Harris song with both a "dynamic female vocal" and a "dynamic male vocal"? The Waterboys' "Fisherman's Blues" is based in a minor key? James Brown's "Cold Sweat" is more syncopated than King Crimson's "Red"? (We're talking Bill Bruford here, who doesn't even know how to drum in straight 4/4 time.)

I had the same bad experience specifying artists, and would also recommend going by favotie songs instead. Seeding it with King Crimson resulted in a lot of mainstream music recommended for its 'mixed acoustic and electric instrumentation,' its 'vocal-centric aesthetic,' and its 'major key tonality,' none of which KC is particularly known for to say the least. Then it suggests a song by Billy Squier [!] in part because of its 'minor key tonality.' (If the tonality isn't relevant, why suggest it is?) But I made a much better song-specific channel based on KC's 'Indiscipline' that had far less mainstream dreck metal, and far more appropriately eccentric selections. It even seemed to choose some based on their unusual 12/8 time signature, which seems to be that song's prominent harmonica part.

And I second the other criticisms: too little jazz and "world" music, and no classical to speak of. And seed material shouldn't show up in the output. The folks at Pandora are also a little pretentious in claiming they evaluate an objective set of criteria. Eventually it breaks down into silly statements like "we're playing this track because it features Brazilian rhythms," which is as much of a punt as saying a particular song has "American rhythms," and reflects the same kind of bias that manifests in social networks.

1/3/06, 11:36 AM

Post a Comment

Subscribe to: Post Comments (Atom)