Dynamic Time Warping allows you to warp signals to compare them.

http://en.wikipedia.org/wiki/Dynamic_time_warping

Even better, you don’t have to implement a good chunk of it! Go ahead and try out Marsyas http://marsyas.info/.

More info: http://rhizome.org/editorial/2011/jul/13/info-mining-george-tzanetakis/

Is it possible to expand this basic idea into something akin to a “tropological music search engine” (or something else with not such an arcane wording)?

For example, say if someone wanted to find all of the songs with a “slow build”, they could give the search engine two or more songs that they thought exemplified for them the concept of “slow build.” An algorithm might correlate each of the parameters between the songs, and find what parameters are most relevant to describing this trope. Another algorithm might then extract a contour that is then used to find other matches in the database.

To take it one step further, the engine might keep track of terms (“slow build” “speeds up at end” “lots of random sections” “very fast”) and allow subsequent users to explore a body of music using the tropes of others.

Of course, such a tool would require expertise on the part of the user. I could imagine someone placing two songs that had little in common, applying the term “groovy”, and getting angry when the engine returns “random” results.

Anyway, I am sure such things occupy your mind plenty already, and my inexperience probably doesn’t begin to grasp all of the difficulties implied by my mutterings.

Cool post.