Impulse Response - Compressors

[Frank Rideau]

Hi there,

I got into reverb impulse responses recently, it's really like a drug ! A lot a cool things out there, I'm thinking of PastToFuture Reverbs for instance.

But what about compressor impulse responses? I've tried an impulse file for the Fairchild 670. It does nice thing to the music, but you really have to use your ears (which is obviously not a bad thing) cause you don't have any indication of gain reduction nor can you adjust input and threshold. You can select different time responses "sample" tho.
But the question is, is this snake oil?

What you guy think of the concept?

[CS70]

Hi there,

I got into reverb impulse responses recently, it's really like a drug ! A lot a cool things out there, I'm thinking of PastToFuture Reverbs for instance.

But what about compressor impulse responses? I've tried an impulse file for the Fairchild 670. It does nice thing to the music, but you really have to use your ears (which is obviously not a bad thing) cause you don't have any indication of gain reduction nor can you adjust input and threshold. You can select different time responses "sample" tho.
But the question is, is this snake oil?

What you guy think of the concept?

Well not sure what you gain. It's possible to emulate compressors pretty well nowadays, in an already compact software format (plugins), so there's little to gain - and quite a bit to lose in flexibility, as you say.

The reason for which reverbs are popular is because real space reverberation is much harder to reproduce computationally. And the reason for which guitar amps and cab emulators are so popular is know to anyone who's tried to haul a 4x12 around - to say nothing of 15 cabs!

That said, isn't it Acustica Audio that uses convolved samples in that way already?

[merlyn]

Processing audio with an impulse response uses convolution -- every sample is multiplied by the impulse response.

This would alter the frequency spectrum of the processed audio but I can't see how it would do any compressing. The impulse response would give the processed audio the sound of a Fairchild but the compression would have to come from elsewhere.

[The Elf]

Let's not forget the Focusrite Liquid Channel, which uses Syntifex's convolution technology (and other tricks) to emulate classic compressors (and is a particular favourite of mine).

https://www.soundonsound.com/reviews/focusrite-liquid-channel

[CS70]

Hm, why do you think you cannot? Just curious. I am not totally sure without giving it some more thought... but in general, convolution is simply a function product; It just happens that both sample set and impulse response are functions over a finite space, defined in a discrete manner as opposite to analytical (at least the response)

The response of a compressor would be the way it reacted to a signal of known shape and level, which exercises every point in the space (say sinusoids) - the tool would have to analyze the signal, normalize it to the known level, apply the convolution for that level and then de-normalize the result of the same amount (it's this last stage that I'm not sure of).

You probably would bake the converter into the image, however, but I guess that happens all the time..

[muzines]

An impulse response is usually a time/frequency thing from a single impulse - it therefore cannot have changing "dynamic" behaviour from a single impulse response, as the single impulse *has* no dynamic behaviour - was only at one level.

Some companies have indeed developed a "dynamic" convolution system to replicate non-linear behaviour according to dynamics, but you have to do multiple impulse response for different levels (and different compressor settings!) and average between them - it's a bit messy.

But no, straight convolution doesn't do much "compression" or anything dynamic - what this impulses are usually doing is giving you the flavour of running a signal through the hardware (transformers, EQ behaviour etc).

[Matt Houghton]

Yeah, as others have said, an impulse response can give you changes to frequency and time. So, effectively, it can capture/reproduce a static EQ curve, and a resonance. Both of which might be features of an analogue compressor.

It can't respond dynamically to an input signal. But some have come up with ways of interpolating between many impulse responses. Sintefex were first to the party, and that technology ended up in the Focusrite Liquid Mix and Liquid Channel. Which to my ears worked well on the EQ side of things, but much less so for gain reduction. The Acustica Audio Nebula and Acqua plug-ins have, effectively, taken that idea much further. They do some great emulations now and in recent years, the compressors have begun to catch up with the EQs in quality. This isn't easy stuff to do yourself: we ran a few workshopson creating your own Nebula libraries back in 2014; it was already complex then, and the search for quality has made it trickier still, so I wouldn't recommend getting into this as a DIY thing!

That said, you'll often find that using an algorithmic compressor in combination with a static IR to give some of the character of a 'preamp' (ie the colouration of the hardware that's not associated with its gain reduction) can do interesting things. But so too does using a clean compressor with any decent character plug-in or hardware!

[merlyn]

Hm, why do you think you cannot?

I was assuming the OP is using the same convolution software with reverb impulses and compressor impulses. A compressor is non-linear so the impulse response (could be a .wav file) wouldn't capture that. It's the non-linearity that people like about analogue gear -- look at tape emulation plugins

[CS70]

Some companies have indeed developed a "dynamic" convolution system to replicate non-linear behaviour according to dynamics, but you have to do multiple impulse response for different levels (and different compressor settings!) and average between them - it's a bit messy.

But no, straight convolution doesn't do much "compression" or anything dynamic - what this impulses are usually doing is giving you the flavour of running a signal through the hardware (transformers, EQ behaviour etc).

Yeah that's a bit the process I had in mind. Since the function space is finite, it's possible to keep it in a reasonable amount of memory. But yes, not something your standard single-impulse response would be able to do - I guess it's similer to what the Acustica people do.

My convolution background come from (admittedly old) days of image processing, not from practical audio, so not entirely familiar on how the term is intended there.. seems more to denote a specific technology or technique here.

[Frank Rideau]

Hm, why do you think you cannot?

I was assuming the OP is using the same convolution software with reverb impulses and compressor impulses.

Yes I am.
For those who are interested to test, there is a link, about 8 rows from the top, for one free IR of the Farichild 670.

https://gumroad.com/pasttofuturereverbs?sort=page_layout

Thanks for the different answers.

[Xavier Lizarraga]

With an compressor impulse response you capture the frequency response for that settings including the contribution of harmonics generated by saturation. But there are ways to capture the "main" IR and "sub" IRs which correstponds to the frequency response for each harmonic and its contribution.

That's how Acustica Audio started and probably the Focusrite mentioned in this thread but I don't have the details. But it doesn't captures the amplitude transfer function for each compressor and also if you convolve you will do that by block so you cannot replicate the dynamic behaviour of a compressor in this way.

Although you use very short block size, classic compressors can be faster than that. I remember to by one of the first Nebula compressors for a 3rd party and it was unable to capture peaks. I hope this helps. Cheers!