You are here

The case for objective compressor tests

All about the tools and techniques involved in capturing sound, in the studio or on location.

The case for objective compressor tests

Postby C.LYDE » Thu Sep 14, 2017 10:25 am

SOS have certainly done a good job of keeping us informed of the various audio compressors available in the market place, many of which are simply to dear for mere mortals to mess around with, never mind owning! Software has of course changed all of that and today most of us have several different types in our DAWs.

The one aspect which I find intriguing is the notion that some compressors are “more desirable” than others for various reasons as mentioned in these articles
https://www.soundonsound.com/techniques ... ompressors
https://www.soundonsound.com/techniques ... r-circuits

For a while now I thought about designing a :geek: test to measure these differences, as I honestly believe, as in most audio purchases – bias is alive and well… :ugeek:

I thought of using Wavelab – 2 identical audio clips e.g. snare hit – one with Colourful (OEM) compressor A the other using WLs built-in compression algorithm. The idea is then to compare FFT analysis of results at various attack, release, threshold etc. settings

Anyone attempt this or have some additional ideas :?:
User avatar
C.LYDE
Regular
Posts: 228
Joined: Mon Oct 21, 2002 11:00 pm
Location: South Africa

Re: The case for objective compressor tests

Postby desmond » Thu Sep 14, 2017 11:48 am

C.LYDE wrote:For a while now I thought about designing a :geek: test to measure these differences, as I honestly believe, as in most audio purchases – bias is alive and well… :ugeek:

You don't need to design a test to prove that, we all know that bias is alive and well! :)

C.LYDE wrote:I thought of using Wavelab – 2 identical audio clips e.g. snare hit – one with Colourful (OEM) compressor A the other using WLs built-in compression algorithm. The idea is then to compare FFT analysis of results at various attack, release, threshold etc. settings

Firstly, what is your hypothesis you wish to prove (or disprove)? That all compressor behaviours are actually the same?

Before doing any work on this, I'd suggest you talk to various people who design or have designed compressors, both hardware and software, to get their handle on things, what choices they made, how do they choose behaviours, or model existing ones, and what they have learnt along the way. I'm sure there's plenty of online research you can do here, there are some great articles in UA's archive, and I'm sure places like Gearslutz as well as other manufacturer/developer sites and blogs.

Behaviour is a complex thing to understand, and it's going to be very hard to do this because the variables are quite sizable - not just how the various time constants behave over differing thresholds, but it's very signal dependent too, especially when you start factoring in "auto" settings. It's going to be quite a lot of "work" to produce any meaningful results - yes, you can run through a few different signals through a bunch of compressors and analyse their output, but I'm reasonably certain it will be hard to prove anything definitive from this, it's too simplistic, and ignores a lot of dynamic behaviour and other modelling such as I/O transformers and other non-linear behaviours.

And while I'm the type that does think it can be worthwhile to do some objective testing to back up the more subjective stuff, these tests often fall down because they are poorly done by people that don't really understand how to test properly to produce meaningful results. And anyway, the *result* of using these tools is always a subjective one anyway - usually the people that want someone to tell them which are the "best" tools to use are the ones inexperienced enough to be able to judge for themselves, and want to shortcut a whole bunch of experience.

I think ultimately, if a given tool gets me where I want to be, that's a good thing, and if there are other tools, with other designs/features/prices/etc that can do broadly the same thing, it doesn't matter. For my needs at least, there are a few "flavours" of compression that have very different characteristics, but I don't really need to have fourteen LA-2A compressor models analysed to see which gives a few per cent closer to a mythical golden model - I just want my music to sound good!

So, for my money at least, the test you propose won't really tell you much, I don't think, and is probably a waste of time - but don't let me stop you if you're curious and want to see what happens... ;)
User avatar
desmond
Jedi Poster
Posts: 7544
Joined: Tue Jan 10, 2006 12:00 am

Re: The case for objective compressor tests

Postby Wonks » Thu Sep 14, 2017 11:59 am

I've done in Cubase with a couple of software compressors using a pink noise source and volume automation to create long and short bursts of noise (also on some software modelled preamps). Just record the results and a spectrum analyser will give you info on how the various frequencies are cut/boosted etc. and the wave amplitude will give you info on the compression and release times.

You can of course use a tone generator as well if you want to see the response at specific frequencies. No reason you can't do it with hardware as long as you've got enough input and output ports on your AI.

One probably with direct comparisons is that some compressors have very vague settings e.g. slow/fast, so duplicating those on a simple software compressor with more accurate setting values is going to require some trial and error.

Also, some standard 'compressors' were designed to be used more as limiters, so have very fast attack times that a basic compressor might struggle to match if it's fastest attack time is say 10ms.
User avatar
Wonks
Jedi Poster
Posts: 5015
Joined: Wed May 28, 2003 11:00 pm
Location: Reading, UK
Now even grumpier than Ivan in his heyday.

Re: The case for objective compressor tests

Postby The Elf » Thu Sep 14, 2017 12:11 pm

On another, less forgiving, forum, a member had made an impressively careful study of the summing of two DAWs and was happy to report that his findings had nulled perfectly, to which he received the reply 'I don't care if they null or not, they sound different'. No amount of objective proof will help dissuade those for whom the technology is simply 'magic'.
User avatar
The Elf
Jedi Poster
Posts: 10885
Joined: Mon Aug 13, 2001 11:00 pm
Location: Sheffield, UK
An Eagle for an Emperor, A Kestrel for a Knave.

Re: The case for objective compressor tests

Postby Matt Houghton » Thu Sep 14, 2017 2:10 pm

I've run some tests out of curiosity. I keep meaning to write up an article at some point... maybe I will when I have some spare time. Meanwhile, I don't have time to go into detail now, but here are a few things that struck me... it's a random list and not comprehensive...

You can't use volume automation to do an accurate test with noise; you need to do an edit (snip and change the clip volumes), so that the level changes instantly when measuring attack/release times. Why? All DAWs ramp the volume over a very short window when automating... which is frustrating for tests in which you wish to measure attack/release times but makes sense when you think about what they're designed to do! And you need to render the noise rather than use a generator, and use the same file to test all the compressors.

When using tones for tests, there's an issue if you wish to use the waveform to examine what's going on visually. You have to zoom in to see what's happening, and that means you need a much higher frequency than the usual 1kHz test tone — I found about 10kHz gave sufficient density of waves at the desired zoom level to see the shape of the attack/release 'curves'. The problem with this, of course, is that many compressors will respond differently to different frequencies, and in audio terms 10kHz is pretty blooming high. Hence, for a general impression, I preferred the noise. It's also worth saying that one of the most interesting things about compression is what happens at the low end — when the timings of the compressor are fast enough to actually distort the LF waveform.

Different compressors' detection circuits have different integration times. In hardware, it's generally fixed (it might offer a peak/RMS detection option). Some software allows you to specify it to the nth degree, but even then, two compressors can have different default settings. So if at first two 'vanilla' software compressors seem to sound different, try matching this parameter, which is sometimes hidden.

The numbers given for attack/release times don't always mean the same thing. By which I mean, it seems that 'attack time' on certain compressors is the time it takes to attenuate by a certain portion of the specified gain reduction... but that certain portion may be different on different models.

It's not all about gain reduction. Don't forget the harmonic distortion of some analogue and analogue-modelling compressors — often this is what makes them characterful and desirable. And don't forget that much of this is level dependent, so it can be quite hard to match things (the 'sweet spot' for distortion/saturation on one compressor may be different for one than another).

For analogue compressors, there's also the noise issue. This can be very different for different topologies. Eg the Neve 2254/33609 diode-bridge design is relatively noisy compared with later FET and VCA designs. But it still sounds nice. Some analogue modelling compressors deliberately add such noise (why oh why oh why I do not know :headbang: ).

...and in practical terms, how easy a compressor is to configure is important. An analogue-style compressor with a handful of controls can be better than an ultra-configurable tool with a million tweakable parameters. And vice versa. Which is why I can love something like the LA-3A and TDR Kotelnikov equally!
Matt Houghton
Frequent Poster
Posts: 725
Joined: Tue Aug 07, 2007 11:00 pm
SOS Reviews Editor

Re: The case for objective compressor tests

Postby CS70 » Thu Sep 14, 2017 2:50 pm

C.LYDE wrote:The one aspect which I find intriguing is the notion that some compressors are “more desirable” than others for various reasons as mentioned in these articles
https://www.soundonsound.com/techniques ... ompressors
https://www.soundonsound.com/techniques ... r-circuits

Bias is not necessarily bad - my guess is that the desirability is due to factors other than the signal-handling functionality alone: interface, learning curve, even the physical feel of the knobs (or their look if software!) have a say. Say the 76 and the LA 2A or 3A and all these classic all share that basic thing, if you know what you want, you can get the result in seconds. Same with a Pultec. It's not only that it sounds nice, but that you can get it to sound nice quickly and predictably and move on.

Always interesting with an analytical comparsion tough, so good luck!
User avatar
CS70
Frequent Poster
Posts: 2241
Joined: Mon Nov 26, 2012 12:00 am
Location: Oslo, Norway
Silver Spoon - Check out our latest video  and the FB page

Re: The case for objective compressor tests

Postby C.LYDE » Mon Sep 18, 2017 9:12 am

The Elf wrote:On another, less forgiving, forum, a member had made an impressively careful study of the summing of two DAWs and was happy to report that his findings had nulled perfectly, to which he received the reply 'I don't care if they null or not, they sound different'. No amount of objective proof will help dissuade those for whom the technology is simply 'magic'.

Any sufficiently advanced technology is indistinguishable from magic. :mrgreen:
User avatar
C.LYDE
Regular
Posts: 228
Joined: Mon Oct 21, 2002 11:00 pm
Location: South Africa

Re: The case for objective compressor tests

Postby C.LYDE » Mon Sep 18, 2017 9:37 am

Thank you all for the inputs :clap:

Definitely some useful insights, some which actually correspond to my train of thought on the subject. I have an appreciation for the non-linear aspects of the various transfer characteristics, but I am curious as how 2 identical themed compressors e.g. 1176 FET style correlate to each other.

So for instance let’s say that “nice sounding” meant high frequency roll-off or mid frequency boost… speaking in simple terms.
Within the heart of these machines (soft or hard) I imagine that the biggest contributor to sound coloration is component phase shift, and whilst it would be difficult to measure – my thinking is that by treating them like black boxes – simply measuring the output for predetermined set of inputs would reveal quite a bit.

Whilst I appreciate that the exercise can get unwieldy, if we were motor enthusiasts chatting about the details of V6 versus straight 4 turbo engines performance – the conversation would naturally include measured performance… and not just – "well it went so fast it blew my toupee off" :lol:
User avatar
C.LYDE
Regular
Posts: 228
Joined: Mon Oct 21, 2002 11:00 pm
Location: South Africa

Re: The case for objective compressor tests

Postby Matt Houghton » Mon Sep 18, 2017 10:36 am

Measuring the frequency response doesn't tell you much if you don't also measure the distortion — ie you must ask yourself whether a bump in the midrange is due to simple gain (like EQ) or to harmonic distortion of lower-frequency elements.

But you need to measure the attack/release times too, to make sure you're comparing like with like...

If you want to measure phase etc, a good cheap way to go about it would be to use Christian Budde's free/donationware VST Analyser. It's a standalone 3-bit only Windows only plug-in, and he does an ASIO VST plug-in — so you can analyse 32-bit plug-ins, or hook your interface up to to the ASIO VST plug-in and to a hardware compressor, and analyse that. Obviously this is measuring the contribution of your interface/converters too, but that should be negligible un this context. It can generate a number of useful plots, including frequency and phase.

I do wish there were a 64-bit and MacOS version!
Matt Houghton
Frequent Poster
Posts: 725
Joined: Tue Aug 07, 2007 11:00 pm
SOS Reviews Editor

Re: The case for objective compressor tests

Postby C.LYDE » Wed Sep 20, 2017 11:44 am

Matt Houghton wrote:Measuring the frequency response doesn't tell you much if you don't also measure the distortion — ie you must ask yourself whether a bump in the midrange is due to simple gain (like EQ) or to harmonic distortion of lower-frequency elements.

In my 1st attempts I'll consider the FFT of the freq. response as inclusive of all harmonic activity - ultimately distortion does exactly that - add freq components.

As soon as I have some decent data - I'll report back...

Matt Houghton wrote:But you need to measure the attack/release times too, to make sure you're comparing like with like...

Totally agree :thumbup: - my kick off point is measure the UAD 1176 vs Waves CLA 76 vs Steinberg Vintage 1176 - will soon discover if all 1 ; 3 ; 5 ; 7 attack times are identical ;)
User avatar
C.LYDE
Regular
Posts: 228
Joined: Mon Oct 21, 2002 11:00 pm
Location: South Africa

Re: The case for objective compressor tests

Postby desmond » Wed Sep 20, 2017 12:07 pm

C.LYDE wrote:Totally agree :thumbup: - my kick off point is measure the UAD 1176 vs Waves CLA 76 vs Steinberg Vintage 1176 - will soon discover if all 1 ; 3 ; 5 ; 7 attack times are identical ;)

Remember there is something like 7 different versions of UA's 1176 on the UAD platform... :headbang:
User avatar
desmond
Jedi Poster
Posts: 7544
Joined: Tue Jan 10, 2006 12:00 am

Re: The case for objective compressor tests

Postby Hugh Robjohns » Thu Sep 21, 2017 10:07 am

I fear you're under-estimating the challenge of what you're hoping to do, and possibly misunderstanding what the measurement tools can and can't do here. The real danger is that you'll get different 'technical measurement' results, but they won't actually tell you what you think they are.

Most tests, for distortion products -- including standard FFTs -- only work with steady-state tones, not with dynamically varying signals. So you could measure the distortion (and frequency response) for a given amount of stable gain reduction... but that will inherently be very low and substantially flat pretty much regardless of the gain-reduction topology.

The real sound differences between compressors are more typically due to their dynamic behaviour (which can also be influenced by the gain-reduction topology and whether they are feed-forward or feed-back designs).

Of course, you can measure and compare attack and release times and slope characteristics (linear, exponential, multi-curves etc) fairly easily, but assessing the audio characteristics during the attack or release phases is very much more challenging.

Personally, I think life is too short to worry about this too much. Once you get to a certain level of competence all compressors do what they are supposed to well, and you simply choose one over another because you like what it does and it allows you get what you want quickly and easily. Same with EQ. Same with Reverb processors, and everything else...

H
User avatar
Hugh Robjohns
Moderator
Posts: 20335
Joined: Thu Jul 24, 2003 11:00 pm
Location: Worcestershire, UK
Technical Editor, Sound On Sound


Re: The case for objective compressor tests

Postby C.LYDE » Wed Sep 27, 2017 11:47 am

My first foray using only the Waves CA76 (stereo). This is really is just a jumping off point.

The sound source in all cases was a sample clip containing a few drum hits - basically something that contained a decent amount of frequency spectra.
Measurement is being done in Steinberg Wavelab using the offline Spectrometer.

My thinking at this stage is pretty basic - static frequency plots using a superposition approach .i.e. constrained to one parameter at a time and assumed linear :yawn:

Without boring everyone with details - some pleasant observations include
- Adjusting attack time (fast to slow)has a observable (measurable) impact on frequency behaviour :)
- The frequency changes were not linear or predictable – quite nice considering this is software, i.e. intention by design and not the idiosyncrasies of electromagnetic behaviour or transistor doping regions. :o
User avatar
C.LYDE
Regular
Posts: 228
Joined: Mon Oct 21, 2002 11:00 pm
Location: South Africa

Re: The case for objective compressor tests

Postby C.LYDE » Wed Sep 27, 2017 11:50 am

Hugh Robjohns wrote:Personally, I think life is too short to worry about this too much. Once you get to a certain level of competence all compressors do what they are supposed to well, and you simply choose one over another because you like what it does and it allows you get what you want quickly and easily. Same with EQ. Same with Reverb processors, and everything else...

H

Ironically we could use that line of thinking with all our gear - but then what use would we have for SOS magazine :lol:
User avatar
C.LYDE
Regular
Posts: 228
Joined: Mon Oct 21, 2002 11:00 pm
Location: South Africa

Re: The case for objective compressor tests

Postby Hugh Robjohns » Wed Sep 27, 2017 12:36 pm

C.LYDE wrote:Ironically we could use that line of thinking with all our gear - but then what use would we have for SOS magazine :lol:

It's not really ironic at all.

There was a time when we differentiated equipment according to their technical specifications because there were big performance differences between budget gear and pro gear.

But that hasn't been the case for at least the last 20 years. Most gear has pretty much impeccable tech specs these days that far exceed most normal requirements.

Where things tend to differ most now is in things like ergonomics, ease of operation, features and facilities, integration with other systems (like DAW automation) and so on....

And thankfully there is still a role for SOS in highlighting these things.

Because I'm old and grey and have the gear and interest, I do still run standard technical bench tests on gear that passes through my hands, but it's pretty rare that I find anything that has poor technical specs.

Yes, there are still differences between budget and pro gear -- a budget A-D might manage an AES17 dynamic range measurement of 118dB (A-wtd) while a high-end box might make 123dB (A-wtd)... but very few would ever be able to hear that difference when anything above 110dB (A-wtd) is essentially blameless in the vast majority of applications!

H
User avatar
Hugh Robjohns
Moderator
Posts: 20335
Joined: Thu Jul 24, 2003 11:00 pm
Location: Worcestershire, UK
Technical Editor, Sound On Sound

Re: The case for objective compressor tests

Postby C.LYDE » Fri Sep 29, 2017 7:28 am

And that's why I believe technical gear discussion should be more than it presently is.

From a recent SOS review

"but although their first compressor is based on a variable-mu tube circuit as used in the Fairchild 660 and 670, it’s not an emulation of those venerable units. Rather, TuCo shares some of their ‘analogue’ sonic characteristics .."

Lets for a second imagine that there thousands of readers (potential or real) that have no idea what this really means... :roll:

We might as well be discussing hi-fi...

From a Whathifi review -
"It lacks a little definition, however, and scenes with plenty of bass tend to highlight the *** uneven tonal balance - it is both bottom- and top-heavy, with the midrange a little recessed and hazy. "
------------------------------------------

Back to my basic sonic finger print comparisons - Waves CA76 vs UAD 1176se

- All parameters identical and fed the same sample.
- The static FFT difference remarkable
- Gain differences especially in the range between 2kHz and 6KHz quite clearly present.
- the notion of more air or presence, or sibilance in one plug-in vs the other would be supported and expected after seeing the FFT.
User avatar
C.LYDE
Regular
Posts: 228
Joined: Mon Oct 21, 2002 11:00 pm
Location: South Africa

Re: The case for objective compressor tests

Postby Kwackman » Fri Sep 29, 2017 8:36 am

C.LYDE wrote:Back to my basic sonic finger print comparisons - Waves CA76 vs UAD 1176se

- All parameters identical and fed the same sample.
- The static FFT difference remarkable
- Gain differences especially in the range between 2kHz and 6KHz quite clearly present.
- the notion of more air or presence, or sibilance in one plug-in vs the other would be supported and expected after seeing the FFT.
To quote yourself...

C.LYDE wrote:Lets for a second imagine that there thousands of readers (potential or real) that have no idea what this really means... :roll:

I'd be in that group, then...
C.LYDE wrote:..The static FFT difference remarkable
- Gain differences especially in the range between 2kHz and 6KHz quite clearly present.
- the notion of more air or presence,...
Again, to quote yourself...
C.LYDE wrote:We might as well be discussing hi-fi...

:D
Enjoy doing your tests, but I just pick one that has controls that make sense to me and tweak to suit. My ageing ears are probably not going to hear the results of "gain differences in the range between 2kHz and 6KHz" after a few tweaks. YMMV!
User avatar
Kwackman
Frequent Poster
Posts: 1021
Joined: Thu Nov 07, 2002 12:00 am
Location: Belfast
Cubase, guitars.

Re: The case for objective compressor tests

Postby Matt Houghton » Fri Sep 29, 2017 10:44 am

C.LYDE wrote:- All parameters identical and fed the same sample.

Have you assumed that identical parameter numbers on each plug-in should yield the same result? If so, that's a flaw in your methodology.

In my experience, the numbers rarely correspond, so you have to work rather harder than that to get the closest match(es) between plug-ins. You have to do it by ear, by watching the gain reduction meter (and the meters aren't always the same... so maybe use a third=party system like the gain reduction meter in PT or Reaper), and by printing the results and zooming in on the waveforms to check that the attack and release are actually doing the same thing. Only when you're sure they're doing the closest possible to the same thing can you really do meaningful comparisons of the results. Then you have to measure different things at different input levels... frequency response, odd and even harmonics, phase response...

In practice, when using the plug-ins, this isn't an issue at all — they're just tools to do a job, and you operate them in the same way until you achieve the desired subjective result. But if trying to come up with some sort of meaningful comparative test, you have to work rather harder!
Matt Houghton
Frequent Poster
Posts: 725
Joined: Tue Aug 07, 2007 11:00 pm
SOS Reviews Editor


Who is online

Users browsing this forum: No registered users