Closed catfact closed 2 years ago
i was finally for some reason able to reproduce using oooooo
on factory norns with latest image. (might try for more minimal repro script.)
possible solution (fixes it for me), but needs a little more refinement:
https://github.com/monome/softcut-lib/pull/64/files
[update] PR below is the fix. issue was internal to softcut but newly exposed by race condition or something.
@catfact , dang, wild to watch how you identified the trouble! is this issue good to close?
I would say so
lately, rebuilding softcut results in sporadic noise after initialization. sometimes it is a loud burst that lasts until the system is stopped. this is happening in latest release image so is now an urgent problem.
bypassing the filter seems to fix it?
https://github.com/schollz/acrostic/commit/179494f69c25740ef916b8d1119a9ebe08a0d19e
so, looking at that:
the SVF buffers are not specifically reset, so added that on this branch; doesn't fix https://github.com/monome/softcut-lib/tree/fix-filter-reset
NaNs/infs could be creeping into input. can test this theory by feeding NaNs into jack, and can address by sanitizing inputs, nbut that has some cost.
can mitigate cost and check stability at same time by considering approximation to
tan
in frequency warping during cutoff calculation; also check final coefficient ranges allowed (looks like all the qway to nyquist which is... optimistic.) (softcut/crone could definitely pick up some cycles by approximating trig functions elsewhere, like for panning.)