Pandora Radio - Listen to Free Internet Radio, Find New...
Rated • 2829 reviews • music • pandora.com
Ah, the tricky business of music recommendation.
Recommendation by genre? Fluid and prone to misclassification, being based on time period and/or general sound, the categories of which are arbitrary and lie on a continuum anyway.
Recommendation by similar artists? Unreliable, considering artists can sound different at various stages of their careers. Evolve or stagnate. Plus albums can sound either really tight ("concept albums") or be a loose mix. Also, if an artist has a huge repertoire, it's difficult to know where to start without being able to listen first, and to listen you must download or purchase, but blindly going into it without knowing which one to get first... it's a cycle.
Recommendation by popularity? Oh please. There is a LOT of great music out there, especially from musicians who put out music independent of major labels.
Recommendations based on usage habits of other users? Just because you like one artist, and some people like a bunch of the same other artists doesn't really say much, and can be way off the mark. It also caters to consumerism, what-is-cool name-dropping, and general pretentiousness.
Enter Pandora, which recommends music based on the deconstructed elements of the actual songs. Fruit of the Music Genome Project, musicians and analysts well-versed in music theory have, for the past five years, spent 20-30 minutes listening for about 400 distinct musical attributes in each song in their ever-growing database - everything from melody, harmony and rhythm, to instrumentation, orchestration, arrangement, lyrics, and vocal harmonies.
So far, I really like it - I've been listing down a lot of songs and artists I never even heard of before.
It seems to work better when you make a station out of a song rather than an artist. During the first song, it goes: "To start things off, we'll play a song that exemplifies the musical style of [artist] which features [list of 5 characteristics]." At the next song, "From here on out we'll be exploring other songs and artists that have musical qualities similar to [artist/song]. This track, [song title] by [other artist] features [list of 5 characteristics]."
And so it seems I have a predilection for:
- mild rhythmic syncopation
- rock influences
- electronica influences
- synth rock arranging
- subtle use of paired vocal harmonies
- surreal lyrics
- use of string ensemble
- major key tonality
- a laid back male vocal
- a breathy female vocal
- thru composed melodic style
- extensive vamping (no idea what that means, but I like the way it sounds! :P)
It's a music stumbler's wet dream come true. You can tweak it by rating songs with a thumbs-up or a thumbs-down. The Flash interface plays commercial-free 128kbps streams. It's mainly a subscription service but seems to also profit also from Amazon and iTunes links. What I like about it is that you get to hear the music right away, no need for third-party clients, and even less need to search-and-download beforehand. And since its analysis is technical, completely unpretentious.
Features I'd like would include an exportable list of songs played and the option to tinker with the "genes" or at least give varying weights to them to give more focus, especially if one likes a specific set of characteristics better than others. I know taste is subjective and very personal, but it's nifty to take an objective approach to pinning down why we like what we like, not to mention s'rillyrilly fun to play with.
Prediction of the future: music genomes will help us compare song structures and single out horribly derivative and formulaic stuff like Nickelback's How You Remind Me / Someday and a whole string of willingly homogenous chart-toppers.
