Duke Listens!: Visit my main blog at MusicMachinery.com

Pandora, the popular music recommendation engine, uses trained musicians to analyze music.  According to Pandora:

our team of thirty musician-analysts have been listening to music, one song at a time, studying and collecting literally hundreds of musical details on every song. It takes 20-30 minutes per song to capture all of the little details that give each recording its magical sound — melody, harmony, instrumentation, rhythm, vocals, lyrics… and more — close to 400 attributes!

Pandora has always been secretive about what the 400 attributes are. That's not surprising since those attributes are the secret sauce that make Pandora work.  I notice however that there is now a Wikipedia page called List of Music Genome Project Attribute that purports to detail the set of attributes Pandora uses for classifying music.  It is a pretty interesting list with about 420 different attributes.  There are over 50 attributes just related to vocals,  22 to guitar, 21 to synth.  There are  a  few non-content based attributes too, such as:

• Production and Lyrics by Rap Icons
• Production and Lyrics by Respected Rap Artists
• Production by a Famous Producer
• Production by an Iconic Producer

Also interesting are the way they classify lyrics:

 There certainly are some puzzling bits here, (such as why are Portuguese lyrics the only non-English lyrics mentioned).

I'm not sure how the wikipedia editors developed this list - it may be that they've just been capturing the output of the Pandora player when it says "I'm playing this song because ..."  or perhaps the editors have some inside knowledge of the genome - as with most entries in the Wikipedia, it is probably best to be a bit of skeptical about the content.