Kazrog said:
So far my initial impression is that the IRs don't sound as "real" without the leading audio before the spike.
Your impression is wrong. No
real speaker delays the signal by 7.5ms, nor does any real speaker need 7.5ms added delay to produce "good tone."
7.5ms added latency is anything
but subtle, and it
hurts, rather than helps, "tone" (more properly, the
accuracy of representation of the speaker's response ), when you do not have the option of convolving with an arbitrarily long impulse response. It is very common for deconvolution algorithms not to preserve absolute time, and yours is obviously among the ones that do not. The initial "silence" is not the result of anything that occurs physically,
it is artificially generated by the deconvolution algorithm. Removing it compromises nothing and improves everything of importance here.
(in particular, the audio processed with non-trimmed IRs has more content at the extreme high and low end of the spectrum.)
No. An IR with fewer data points (the effect your leading silence is causing) has
less detail
at all frequencies. In order to do an
accurate analysis of an IR, you must limit the spectral conversion (FFT)
to the total number of samples that will actually be used in the target IR. In the case of the Axe-Fx, the number is 1024 points in "Hi-res" mode and
512 points in "Lo-Res" mode, and the IR will begin with the first data sample if you use AlbertA's free converter. When you've artifically added 7.5ms of leading silence, "Hi-Res" mode becomes 667 points and "Lo-Res" is
152 points. We had numerous discussions on the old board about the audibility of the difference between 512 and 1024 points (which we had to do "creatively" prior to Cliff's giving us the "Hi-res" option). Everyone who made the comparison agreed that 1024 points sounded noticeably better. What do you figure are the odds that these people will perceive 152-point IRs as acceptable? That's what you're offering them now.
Obviously, even a very small amount of latency like we're discussing is best avoided, but not at the expense of tone.
If eliminating it comes at the "expense of tone," then you're doing something badly wrong.
We're figuring out if there's a way to reduce the latency without compromising tone,
Yes, there is.
Edit out leading silence, and thus shift the entire impulse response backward in time, where it resides in the physical world. The only effect on spectral content will be due to the added data you now have within the IR window, but that is a
good thing.