Difference in sound with Impulse plugins

[Arksun]

Something I'm curious about with regards to what's going on 'under the hood'

I was just comparing how an impulse response taken from a church sounds in two different IR plugins, SIR and Voxengo's AFImpulse.

With all the settings set to flat (ie no extra pre-delay, use of eq, stretch, env etc in SIR) in theory the results should be the same but they're not.

I've posted up a couple audio examples to show this. A snippet of vocal ( copyright 2006 Luminary ) i processed.

Both examples are 100% wet. 24bit 44k wavs, using the same impulse.

The AFImpulse one sounds less bright than the SIR one to my ears, kinda smoother, but then the SIR sounds a little more spacious.

I can't decide which I prefer, but it is interesting that there's such a noticeable difference. Has anyone compared how the same impulses sound in other IR plugins and have a personal favourite?

Here's the link for the .wavs:

http://www.musicprogressive.com/impulse/

[93143]

Is the impulse response at 44.1 kHz too? I've heard that SIR has a less accurate resampling algorithm than Pristine Space (and presumably AFImpulse). Interestingly enough, it doesn't result in a definite degradation in this application, and some people prefer the 'shine' it gives certain impulses...

There could also be a difference in the FFT shortcut used for convolution (real convolution is too CPU-intensive, as this MATLAB user can assure you), but I'm not sure about that. Others here know more.

[shamann]

Could be also be a lot of things. The final output might not simply be the straight convolution process, there could be a hidden filter or limiter or other things going on before you hear the sound. A similar example: load a wave file in several soft samplers and compare the output. They don't all sound the same. Some will playback quieter, some use steeper anti-aliasing filters, some perform different interpolation on transpose, and so on.

With convolution engines, windowing could very well be performed differently, that would make a difference in the sound for sure.

[Arksun]

The impulse response was only a 16bit, 44K .wav yup.

SIR v1.011's latency is 8960
AFImpulse was set at its default 4160 latency

The only plugins used to make those .wavs were the Impulse Response plugin itself, and a UV22 HR plug set to 24bit dither

So the differences are very much down to whats going on inside those Impulse response plugs.

Makes me want to try other IR plugs now to compare.

[Debutante]

Heeey!!! that is pretty wierd. The one from SIR sounds even a little fuller, or more robust or something? But the other one seems to have a louder low ambience component than SIR, like a little more ring. Does anyone else hear this?.... or do I exist in my own little funkydom

[Freeform]

Upon zooming in I noticed that SIR has many more vibration variations than the AF Impulse. I'd compare it to the original dry sample & find out if those vibrations are generated, or if the shaping is technically weaker. Interesting however... My guess is since there is a large amount of multiplication & averaging going on in convolution, they are using different formulas. Nice ear.

[shamann]

There's no question that they sound different. Really noticeable in the high frequencies, and the overall volume/loudness isn't the same.

But my whole point above is that a convolution plugin isn't just a straightforward representation of the textbook maths involved. There are all sorts of variables that can be added/subtracted/altered that will always affect the output in different implementations.

[Debutante]

I think as long as you like the sound it's fine. The only problem I can think of in this case is maybe trying to accurately represent a particular room as THE core facet in what you're doing, but for most of us here, I imagine these differences are just fun

[Freeform]

But my whole point above is that a convolution plugin isn't just a straightforward representation of the textbook maths involved. There are all sorts of variables that can be added/subtracted/altered that will always affect the output in different implementations.

That's a very nice point as well. I am not a big fan of static internal effects/processing w/ no bypass option. Especially when it's a limiter/peak-editing effect. Along other lines, some people use antialiasing formulas, and others bandlimit, and I've even tried a basic softsampler that I *swear* had reverb in it. Really though, it's what makes each vst/vsti unique I suppose. A minimoog clone made by Arturia will never sound identical to a creation by another team even though they may have created from the same schematics. This is due to the fact that there are several ways to make a signal do the same *basic* thing (including mathematics) , but the collective effect of each being slightly different than the other would lead to a total effect that sounds much different than the other.