Indeed, the new IR's are fantastic in this release. Incidentally, I am using S-gear as my IR loader for any amp sim that I use. Someone pointed out on another thread that, for some reason, S-gear's IR loader sounds really good. I think I know why...
Mike opted to allow 20,000 samples in a long impulse response. As a reference, when you use a 500 ms long IR at, say, 44.1 kHz, you, of course, have about 22k samples. Most other IR loaders for guitar cab sims limit the number of samples to far less than this, because most of the tail of the IR is close to zero in value. Seems reasonable, right?
However, I think they have made a BIG mistake. IR loaders implement the requisite circular convolution using an FFT / IFFT to provide zero-latency convolution. Therein lies the rub - if you cut the IR short, you limit the frequency resolution in the FFT. In fact, it is often common to zero-pad signals to increase the frequency resolution when taking the FFT. So by keeping more of the IR, even if it is close to zero, you increase the frequency resolution of the result!
So please, Mike, don't cut the IR's shorter, even if the tails are zero or close to zero, in future releases. Whatever FFT implementation you are using is quite efficient, and it isn't over-taxing a modern CPU. I will note that it has been a while since I dug deeply into signal processing theory, so I may be off the mark here - if someone knows more please feel free to correct me... But I think I am on the right track. I recently implemented a zero-latency convolution for a real-time systems course that I teach where my students implement some audio effects, so I spent some time looking at IR loader implementations.
