That was great.
Wish the explanation video was longer to get a better sense of what is going on.
Thanks @rodrigo.constanzo. We were given a very stringent time limit, and I wanted to make sure I got clips of all you people rocking it hard from last November into my 90" (NIME was even terser: 30" second intro!)
Oh, and I should tag @d.murray-rust as well…
Gotta have that director’s cut!
Is your setup for this (roughly) an expansion of that “downsampled novelty curve to find bits and resample them” approach that you showed at one of the plenaries a while back?
I saw someone (@d.murray-rust?) was using some real jittery novelty curve stuff along the same kind of way that looked interesting.
It’s not unrelated, because I only really have one idea. This one is decomposing the input using NMF, and buidling a library of the decomposed snippets, then doing some pretty bog standard KDTree matching to resequence decomposed bits of its history. The thing I showed at that plenary was conceptually similar, except that the event detection was done a multple timescales using the error signals from neural networks trying to recreate their input.
Dave was (is) indeed, making some cool gizmos using novelty to automate sampling off other the two of us.