No... I'm not absolutely sure I understand what you're describing. I'm not sorting audio for the sake of tagging it but as the first step describing all of it's perceptual qualities as they change over time to create a template that can be extrapolated to any other part of that multi-parameter matrix, for instance if there's no "D" note sampled, but there's 10 layers of velocity and three playing styles then the appropriate pitch, spectral, and other decays can be modeled and re-sythesize a new version for each step of the matrix: so when you play "D" each envelope matches analyzed curve for every velocity (how hard the note is struck or played or whatever), and for every velocity of every detected playing style, and on and on. It represents, usually, thousands of nuanced decays as you call them...
Which is interesting on it's own, but more importantly all this data can be abstracted as automation (which at the moment we'd use MIDI for, but which is obviously woefully inadequate). Um, ok I can't think of a way to put this in civilian terms, but maybe I'll be standing around to explain it sometime.
I don't drive (in fact don't know how) but I'm sure I'll figure out where the factory is and bus over sometime, I'm still hopinh this Thursday. I'd love to collaborate with hardware people somtime.
i'd look at actually extracting the waves with and processing data points on the file rather than the gross midi. i know there's some freeware fft programs that would help. then you're going to want to look at what the wave is doing in a variety of parameters, envelope, pitch, harmonics, etc. some of the earlier tracker software was doing some pretty good pitch bending without altering duration, if you can dig up some of their algorithms you may start with an axle instead of a boulder for this wheel.
what part of town do you live in? there's a chance someone might be willing to slow down enough for you to dive in the back window on their way to the factory