Re: music genome project

http://pandora.com/

I have used this. It's fun, and it has actually suggested music to me which I went out and bought.

That said, the music genome project seems to classify music on strictly formal grounds -- male or female vocals, slow or fast rhythms, etc. Which is fine and good (I'm quite the formalist myself, as any of my cultural studies friend will attest), but it doesn't give the system any ability to distinguish between two bands producing similar-sounding, similarly-structured music, where one piece of music is good, and the other is crap.

Of course, most people can't make this distinction, either, which accounts for the popularity of openly derivative, openly crappy content such as, say, Candlebox, or the Monkees.

Reply from a friend:

Yes, what it boils down to is that Pandora - and any other service or person - can only suggest tracks based on measurable well-defined properties. Whether it is good or bad music will be a matter of personal taste (with a few undeniable exceptions).

And sometimes Pandora is just plain wrong: I entered Goldfrapp as the search term, and Pandora came up with songs by Ace of Base and Belinda Carlisle!

My response:

I was being a bit tongue in cheek -- of course value questions such as "good" vs. "crap" are subjective. However, there are ways that the system could try to use "quality" (or at least the chronology of musical variation) as one of its axes, as it tries to guess what to play next.

• Get album or artist ratings from the web (Amazon, emusic.com, or elsewhere), using them as a raw index of "quality".
• Text-mine music journalism archives for positive reviews, possibly weighing these against the popular opinion. Perhaps the system's profile of the listener should include the listener's likelihood of agreeing with the Amazon crowd, the eMusic crowd, the Pitchfork crowd, the Rolling Stone crowd . . . .
• More interestingly, do some more detailed language processing on the text of all those reviews, looking for particular language, such as sentence forms built around verbs "inspired by", "derived from", "echoes", etc. From that, the system might infer that certain music emerges from other music. Whether the listener prefers the early example of something (Television, Velvet Underground) or later examples of something similar (The Strokes) could also be part of their profile.
• By default, the system could try to give priority to earlier examples of a particular form, or form combinations, such as close female harmony/electric guitar. Unless a later example of that formula is clearly highly-rated (using some of the above metrics), the earlier stuff should be weighted more heavily, since it is probably more original music, and perhaps of higher overall musical value.