Closed beserge closed 3 years ago
Converted to draft while we provide additional fixes.
as a brute-force fix to the actual crash issue we can:
if (low_ != low_)
low_ = 0.f;
for low_
, band_
, high_
, peak_
, notch_
but that's not very efficient being called 48000 times a second. So, it'd be ideal to find what is causing NaN in the first place.
anything that would end up with 1 / x
where x
is NaN, infinity or -infinity would result in NaN (or multiplying against inf/-inf)
Based on my experience and the notes in the #72 this could be more common on signals with big transients square waves seem the most common culprit). Though, you mentioned seeing it crash with a steady input of just 1.0f
.
Fixes the crashing bug by making resonance work internally from .05-1. A better fix should be coming.