I am just passing this information on. It is in this latest issue of Audio Technology written by Mike Stavrou a well known engineer here in Australia.

We all have a collection of digital plugin compressors. And why is it that some might impress us and others not so much. This is a test that will find out which compressors are best for use in our productions.

Music compression is all about the shape of its movement. This can be difficult to hear. There is the dynamaic movement of the compressor and artifacts that are also created. This test exposes the inaccuracies and strange anomalies in bad compressor designs. While the music is present and it is constantly moving up and down it can be hard to hear the volume changes that are attributed purely to your compressor. To eliminate any confusion the idea is to eliminate the music leaving behind only the volume shaping the compressor is performing.

What to do:

Find the right source. A single instrument or a punchy complex dynamic uncompressed mix is also good.

Insert this on a track. And copy it to an second track. Invert the phase of the second track and adjust volumes for a complete null. Faders should be at unity.

Insert test compressor on second track. Perhaps first set the compressor so it is not doing anything. (eg high threshold low ratio etc but still inserted and operating) Check for null. A small delay in the plugin will prevent a complete null. If this is the case then try advancing the second track a few ms or so and see if that improves it.

What you will hear:

Each time the compressor grabs hold of the sound and reduces its gain the source will jump into view. You should not hear and clicks that are not part of the music. Any clicks farts or noises generated by this particular unit are not a byproduct of the compression itself. These strange noises, thumps or distortion are being added by the workings of this compressor and will appear in your mix.

If you use this same compressor plugin on 6 individual tracks than these things multiply as well. Some compressors are better than others. Instead of smooth or quick ramping levels you might hear the swell made of jittery steps. If it jumps all over the place it is bad. You can smooth this by changing release time (200 ms or more)

It is a great test to do on demo versions before you invest in some expensive compressor plugin. Test all the ones you have. You can use this test to also setup a compressor in mastering as you normally would then do the null test to see how it effecting your audio. Then fine tune for better results in the null state.

What to Aim for:

You play with the controls to find the most elastic setting where the sound rushes towards you and away with a spring in its step. Try to get the movement to compliment the rhythm of the source. That is when compression feels the best. When it bounces along with the energy and the vibe of the track. Then the compressor becomes the finishing touch to your production instead of the first stage of accidental destruction.

Take turns inserting each of your compressors and limiters into the second track and hear with clarity the vast differences between them. Expensive analog compressors have a smooth rubber band like action. But does not mean we can't get our best digital plugins to do the same.

Excellent suggestion, Jeff. Thanks!

Such a test could also reveal distortion and aliasing, at least if they were actually audible in isolation. There would, however, still be much about the compressor's behavior that such a test would not reveal.

For example, I have an article here (sorry, no link: unfortunately I did not note where I originally found it) explaining why different hardware 1176 units sound different from one another. Turns out it was the shape of the release curve that varied from one unit to the next, due to component tolerances. A null test would have indicated that a difference existed, but would it have been obvious what that difference was?

Another caveat is that compressor settings would be critical for comparing two plugins.

For example, a compressor that offered an effectively instantaneous attack time would be more prone to artifacts than another that only went down to 10ms, so any valid comparison would require them to both be set to the same attack time.

And setting up two different compressors identically could prove difficult. What if one had a threshold control and the other had an input level control? Or one offered an adjustable knee and the other didn't?

Sounds like an interesting exercise.  

Yep.

Another thing to file away and experiment with in my 'free' time

Listening to a difference signal and trying to draw any subjective conclusions is never ever a good approach.

Null tests are a very good way to establish whether 2 signals are different, and to what degree. And you can potentially analyze the difference signal to, for instance, see if it's random noise or restricted to a certain frequency range or something like that, or by comparing it in microscopic detail to the original signal, or by using test signals.

But you can't really get anything out of just listening to the difference signal on its own because you don't know whether what you're hearing was added to or taken away from the original signal. 

And you can't tell whether what you're hearing is even audible in the context of the original signal, particularly if the difference signal is amplified (as it usually is). Some people may want to argue this, but, IMO, it's a terrible argument - if the differences are so easily audible, why bother with listening to the difference signal at all?

Thanks guys for reading this. A few things. I have been doing some testing and it seems it is quite revealing. I would like to address a few of Dave's concerns.

Firstly this is a good test for digital plugins. Most analog units tend to behave in a more elastic way. Differences in the 1176 release setting would have manifest itself as how the sound jumps at you (similar for several units) and then goes away (different for several units) etc.

Stav uses this analogy at the start of the article. Imagine a stage populated by a collection of animated characters - each one jumping forward and back while expressing their thing. To reduce the overall dynamic range each one should be held back with a rubber band tensioned specifically for thier range of movement: a moderate 'jumper' with a medium tensioned rubber band, a hyperactive jumper with a tight rubber band and so on. The tension equates to the ratio setting on your compressor.

This example above might be looking at how the individual compressors are behaving on tracks. A total mix going through our mastering compressor is obvioulsy a single rubber band if you like over our total mix.

Compressor settings do not have to be critical to compare one plugin to another. What you are listening for is how jerky the rubber band motion is when any given plugin is doing its thing. If it is jerky and stepped and also a bit wild as Stav points out that is not good. Try adjusting release among other things to see if you can smooth this movement out. The better plugins can give you a smoother more bouncy elastic type effect each time the compressor grabs onto the audio. This is the main thing you are listening for. Because that is how the compressor is applying its gain reduction when the music is back there and you are hearing being ducked away instead of nothing or silence and burts of music suddenly moving toward you. The null test is exposing how a digital compressor plugin is behaving and it can tell you a lot.

Watch out for the delays that some plugins are introducing. You must aim for a complete or close to null before you go any further. You need to be savvy with track advancing in time and by very small amounts as well. You can try delaying the track without the compressor on it as well. I have had success both ways.

I setup in both Sonar and Studio One for these tests. You do have to tweak the volume of the track with the plugin inserted for finer null tuning for each plugin. You don't have to alter track delay or advance settings very often either. When I do this I do it in Studio One as it is easy there.

I found that certain compressors performed well. The standard Sonitus compressor did well. The Sonitus multiband not so well. The LP64 Multiband EQ also did very nicely. Could get to bounce in and out very well. A few other compressors and limiters did not do well, especially the limiters. Very jerky response. Won't use them again after hearing them in this situation.

It is interesting when you do get something like the LP64 working nicely it is amazing how much the attack parameter effects things in terms of that bounce feeling. Slowing it down means the compressor is not doing so much and the result is mostly quiet. Speed up attack and you start to hear it more often. It is very interesting playing with all of the (compressor) parameters without the music being present most of the time.

Other things show up some interesting results like tape simulation. Some of them added some not so nice distortion and artifcats. It exposes what your plugins are adding.

EQ's are interesting. When you effect a band (boost or cut) you only hear the frequencies in that band. You really get to hear what parts of the spectrum you are effecting with eq.

Drew makes a good point, which is that you can't necessarily assume that what you detect in the difference file is actually an audible artifact, much less an unpleasant artifact. The null test tells you that a difference exists, but not a whole lot beyond that.

I am reminded of a long thread on either Gearslutz or KVR, I forget which one, where compressors were being evaluated based on aliasing measurements.

Somebody had discovered that some compressors showed aliasing, while others did not. Everybody went nuts taking spectral pictures of their compressors and posting them. And they all read WAY too much into those measurements. Nobody seemed to notice that the aliasing was much too low to ever be heard.

I have found with digital plug ins that the attack and release is always a clouded affair. Mild attack and release seem to work fine but most break down after even a 4:1 ratio and around -6 gain reduction. They just sound smeared past that point. One notable exception is the Precision Buss Compressor from UAD which can be pushed farther than most. I have not found even one that will emulate an 1176 in all button mode. Most get really smeared when trying to duplicate this sound. There is not the tight grab and slow clear release that hardware provides.

Another challenge is the depth & weight of the sound that transformers impart. A lot of plug ins don't map to the real world counterparts in the low mids and bottom end. That said I agree that the Sonitus is a good digital compressor. Neutral and effective but unfortunately not enhancing which is sometimes a desire. Overall in my experimentation, digital compressors, when applied throughout a mix as you would real compressors, give a haze and cloudiness to the sound which at first may be appealing but when compared to commercial music makes it obvious that transparent compression in hardware is a better approach.

Stavrou's book (not cheap) is a gold mine of recording and mixing knowledge.  His earlier methodology for setting compressors is a very valid one, and it focuses on setting the attack time to get the desired hardness and the setting the release time to enhance the groove.