Gain Transfer Function Measurement Tool ?

[kylen]

This isn't really an effect but a tool to help me characterize my compressors and limiters (maybe) by seeing what the transfer function or gain reduction curves look like. Many manufactures publish these curves (especially hardware folks) many don't.

At any rate I thought there might be a software plugin I could put in some host and use something like this in real time:

Transfer Function IN Tool -> Compressor -> Transfer Function OUT Tool

That's just one possible implementation of the tool - some developers here like Wide Boy and Voxengo already do communication betewwn plugin instances and have sender/receiver type connections so it can be done.

Anyway I want to look at the 'actual' transfer curves of my plugins and try to see if a particular curve is responsible for a particular character or coloration. This might have something to do with future dynamics convolution processors so maybe I'll look there too.

I'm not a DSP guy so I'm not trying to write code here just look at transfer curves in real time (or could be a batch mode operation I guess too). Any ideas ?

Thanx

[kylen]

Well I've spent a lot of today lookin around and I just don't see a tool for this so I assume I must be an idiot.

I do see compressor gain reduction curves like this one:

http://www.manleylabs.com/PDF/PRO_Manua ... 0chart.pdf

And there are compressors that show what the transfer curve is you currently should have, like the GCO-1 from Kjaerhus:

http://www.kjaerhusaudio.com/images/gco1.jpg

But I don't see any tools available that will measure the actual compression gain reduction curve. I must be missing some information about calibration or testing of compressors obviously.

Any Ideas ?

[C00kie]

You can create such diagrams yourself, either by calculation or by measuring.

Measuring: all you need is a test tone generator and/or an audio recording program. Some audio editors such as CoolEdit feature a tone generator. Make a tone that gradually goes louder & louder, in a linear way on the dB scale. Then feed that through the compressor and watch the output levels.

Calculation: iIf you know the threshold and compression ratio set at the compressor you can draw the curve yourself also: straight line at 45 degrees going up until the threshold level. There the curve bends (account for "soft knee") and divide the angle by the ratio. A ratio of 2:1 means an angle of 22 degrees, ratio of infinite means a flat line.

With a compressor set at different settings, you will get different curves. The curve is just a visualisation of what the output level is at a certain input level, not accounting for attack & release times.

What to do with it, I truely don't know. Stick on the wall above yor bed, or impress friends...

[kylen]

straight line at 45 degrees going up until the threshold level...

Thanks for your comments C00kie - that's the part of the actual transfer curve I want to see the most (the next is knee shape). Where a lot of compressors have a flat diagonal 45 degree (1:1 ratio) transfer curve up to the threshold (not counting influence of knee) I think that some also have a different shape or ratio in that range (below threshold). That's what I want to see so I can catagorize my compressors into ones that are very transparent (flat 1:1 transfer up to threshold) and ones that add meat and body and smoothing in a pleasant way (I think this is where the transfer curve is not flat up to the threshold - maybe slightly parabolic or has less that 45 degree angle).

At any rate the old Cakewalk FX compressor has an adjustable transfer curve you can add 9 points to, Cool Edit/Audition has a dynamics processor where you can add 6 points to the transfer function, Neodynium lets you add 4 points (with independent ratio, attack/release) although it doesn't display the expected transfer curve.

Here's an article by Rip Rowan from his old compressor shootout that shows the multi-point transfer curves for folks that are wondering:
http://www.prorec.com/prorec/articles.n ... enDocument

Anyway, as you mentioned C00kie I'll probably plot out a couple by hand to see if the inherent transfer curve of a compressor is worth characterizing. I've got Audition so I can generate some useful tones I think (maybe there's some standard tone type manufacturers use to generate the gain reduction curves).

I'm going to guess before I do this and say that a transparent compressor like Polysquasher and Soniformer have flat transfer curves. Beefy and smoothing compressors like the ones in Voxformer and Crunchessor may have something other than a flat (1:1) transfer ration up to threshold (not counting knee).

[kylen]

I plotted out some of my compressors in Excel and discovered the beefy and coloring ones (Tracks VST Comp, Voxengo Voxformer, Crunchessor) continue to compress below the "threshold". These don't actually have a threshold control. There is a point on the transfer curve where the ratio goes 1:1 so I guess I could loosely call that the threshold point(?). These kinds of compressors either have a single knob or drive control where you can tune around for the sweet spot and get them sounding really good for certain types of jobs. The compressors with these types of compression curves seem to be able to handle peaks and smooth out the body of the track at all levels. They smoothed out some pretty tricky vocal tracks anyway.

The really transparent and non-coloring compressors seem to be more like limiters acting as compressors (Polysquasher, Soniformer) as they don't do anything below the threshold but don't have large limiting ratios. These have an actual threshold control that you can tune in if you need a tool to deal with peaks or audio above a threshold. Maybe good for rock guitar - something where the body of the sound is already consistent. Possibly using compressors with these types of compression curves would also allow low threshold and low ratios so they can deal with more of the body of the audio.

But these two compressor types just don't sound the same to me - even though I can drop the thresholds down low on the 2nd type. I've still got some experimenting to do it looks like. In the case of Polysquasher it might be that using a huge knee (it can go up to 40dB) might do the trick and adjusting the threshold way low for the sweet spot. Hmmm time for more plotting !

I'm not saying any of these compressor types is better than the next one - I'm just trying to get better at choosing the right tool for the right job - and be quick about it!

[citizenchunk]

I'm not saying any of these compressor types is better than the next one - I'm just trying to get better at choosing the right tool for the right job - and be quick about it!

in the words of Charley Brown, "good grief!" if you want to find the right tool for the job, use your ears, not a graph! transfer function graphs won't tell you ANYTHING about a how a compressor sounds; all it will tell you (as you have correctly pointed out) is at what input level it begins to affect the sound. but that and "sonic transparency" are 2 completely different things.

the heart and soul of any compressor is its SIDECHAIN--more precisely, how the sidechain level is detected/averaged, which are time domain processes. that is what gives a compressor transparency or color.

most feed-forward compressors have linear gain reduction, meaning that the gain reduction is linear (in dB), hence predictable, above the threshold. by feed-forward, we mean that it senses the sidechain signal at the input, rather than the output.

compressors that exhibit a non-linear gain reduction (as you have noted in the manley manual and in tests) are "feed-back" compressors, meaning that they sense the sidechain level at the OUTPUT, after the gain reduction. this makes the gain reduction nonlinear. they are also "nonlinear with respect to time" (for lack of a better phrase), meaning that the current gain reduction is dependent on previous inputs, which are dependent on previous gain reductions, etc. as a result, it is difficult (if not impossible) to pinpoint an exact threshold. so these compressors effectively have no exact threshold, but rather sidechain gain control.

the VariMu, Voxformer and Crunchessor are all feed-back compressors, which accounts for their nonlinear compression.

then of course there are feed-forward compressors with nonlinear soft-knee transfer curves. some of them (like mine) start to compress below the threshold in order to ease into the compression ratio above the threshold. some do not compress below the threshold, but still ease into the ratio above the threshold. these are called "easy-over" soft-knees. in terms of design, there is no right or wrong, just preference.

it should be noted, however, that even though these feed-forward soft-knee compressors have nonlinear transfer functions, they are usually still "linear with respect to time," as the compression is independent of the preceding inputs.

but none of this theory will tell you which compressor is the right one for the job. only your ears can do that. so just try a little listening first.

take care.

== chunk

[Aleksey Vaneev]

citizenchunk, just a clarification - Crunchessor is not a feedback compressor (i.e. it's not like Voxformer).

[citizenchunk]

oops! sorry Aleksey.

hey, now that i think about it, that's a little odd. isn't it? i thought you were going for a classic valve/opto feedback comp with that one, and that the "Drive" param was for sidechain drive. or is the Drive actually driving the input against a preset thresh/ratio? is it using some sort of static soft-knee algorithm? well, whatever it is, it works very well. i messed around with the demo back when it was just released and found some of the colorful sidechain settings, along with the valve emulation, very, well, colorful. cool plug.

== chunk

[kylen]

in the words of Charley Brown, "good grief!" if you want to find the right tool for the job, use your ears, not a graph! transfer function graphs won't tell you ANYTHING about a how a compressor sounds; all it will tell you (as you have correctly pointed out) is at what input level it begins to affect the sound. but that and "sonic transparency" are 2 completely different things.

Well you seem to be excited about this - that's good I am also. After listening to compressors for many moons and deciding some are spongier and more transparent than others I've decided to take a look at a few characteristics to see why that is - who said I wasn't using my ears? In other words why do they sound different to me, so I'm listening to the device OK? It seems to me that the effect the compressor has on the input signal and to what degree the transfer graph is linear or non-linear is part of it. You mention the side-chain has a great deal to do with it too - I agree also.

the heart and soul of any compressor is its SIDECHAIN--more precisely, how the sidechain level is detected/averaged, which are time domain processes. that is what gives a compressor transparency or color.

Yes - you've said this in other posts. It would seem that this very directly affects gain reduction thru the device. One way to represent this is the gain reduction transfer curve.

then of course there are feed-forward compressors with nonlinear soft-knee transfer curves. some of them (like mine) start to compress below the threshold in order to ease into the compression ratio above the threshold. some do not compress below the threshold, but still ease into the ratio above the threshold. these are called "easy-over" soft-knees. in terms of design, there is no right or wrong, just preference.

Yes this is somewhat like the shape I'm seeing on the transfer curves below the threshold - or unity ratio (1:1) in the thresholdless compressors. But the knee and continuing curve are not at the 1:1 portion of the transfer curve. The transfer curve actually has about a 1.19:1 ratio at very low dB levels (< 60dB). It appears to be some smaller section of a parabolic curve that crosses unity ratio at about -8.6dB (for a 0dB 1KHz test tone withe 20dB drive in Voxformer).

but none of this theory will tell you which compressor is the right one for the job. only your ears can do that. so just try a little listening first.

Well I disagree of course - and you seem to be outraged that someone is lifting up the skirts of a compressor to see what's underneath. Yes - let me continue to listen, I will - thanks, that's great advice, who said I stopped listening? I was hoping to talk to a few compressor developers to see what they think. Maybe there are some others that don't take this so personal.

I'm still wondering how nice it might be to get various transfer curves - which are just the representation of an aspect of the gain reduction of a compressor - in the same compressor. The older Cakewalk FX compressor has it. In other words a threshold, possibly a knee, but also the ability to tilt the entire curve at some unity or pivot point. So now that I've seen a few things and listened to a lot more maybe I'll see if the compressors that allow user configurable transfer curves can be made to create the sounds I need for a particular job.

I wouldn't mind being able to hear the results of poking around the transfer curve of Polysquasher sounds or even Soniformer - but in the case of Polysquasher at least I could try to simulate it with a deep deep threshold and knee but light ratio 1.18:1 or thereabouts.

Squishy, transparent, leveling, smoothing. Those are some of the sounds I need from compressors for various jobs.

[citizenchunk]

hey man, relax. i didn't want to "get all up in your grill" or nothing. honestly, i'm not that excited/heated about this. and please forgive me for assuming you weren't listening; i shouldn't have jumped to conclusions. but i should clarify some details...

the heart and soul of any compressor is its SIDECHAIN--more precisely, how the sidechain level is detected/averaged, which are time domain processes. that is what gives a compressor transparency or color.

Yes - you've said this in other posts. It would seem that this very directly affects gain reduction thru the device. One way to represent this is the gain reduction transfer curve.

not really. as i said, the sidechain is a time domain process--amplitude with respect to time. a transfer curve (if we're using the same terminology) represents output amplitude with respect to input amplitude.

but none of this theory will tell you which compressor is the right one for the job. only your ears can do that. so just try a little listening first.

Well I disagree of course - and you seem to be outraged that someone is lifting up the skirts of a compressor to see what's underneath.

heh heh... not really. i try not to get emotional on this forum. actually, i applaud your interest in this stuff. honestly, i was just trying to be helpful. and when i type, i type a lot, which may come off as excitement.

then of course there are feed-forward compressors with nonlinear soft-knee transfer curves. some of them (like mine) start to compress below the threshold...

Yes this is somewhat like the shape I'm seeing on the transfer curves below the threshold - or unity ratio (1:1) in the thresholdless compressors. But the knee and continuing curve are not at the 1:1 portion of the transfer curve. The transfer curve actually has about a 1.19:1 ratio at very low dB levels (< 60dB). It appears to be some smaller section of a parabolic curve that crosses unity ratio at about -8.6dB (for a 0dB 1KHz test tone withe 20dB drive in Voxformer).

you have to remember, VariMu and Voxformer are feed-back compressors, which are very different from feed-forward designs (which i was describing in the quoted text).

i should clarify my description of feed-back designs: a feed-back design compresses whenever there is any sidechain signal at all, and the gain redux is proportional to the sidechain level. theoretically, if the sidechain is absolutely dead (-inf dB), there will be no gain redux--complete linearity, input to output. however, we both know that theory is one thing and implementation is another. analog components never react ideally, which may cause a tiny leak into the sidechain (proportional to the input), even when the Compression knob is all the way "off". or perhaps a bad photo-cell will perceive a sidechain when there is none. in the digital realm, sometimes a design doesn't actually have an absolute "off" position (excluding bypass), and is always "leaking" sidechain level. and sometimes programmers just get sloppy (like me, on occaision ).

also, this is disregarding the effects of the I/O transformers or, in Voxformer's case, any "warming" processing that may or may not be going on. there are many ways to explain the behavior that you are noticing. but IMO, a transfer graph won't tell you the whole story.

if you really want to peak under the hood of a compressor, here's a little experiment that might shed a little more light on the subject:

1. create a file of 10 kHz sine wave, at -20dB
2. edit it such that it periodically changes gain to 0dB for increasingly longer durations. e.g. 5s -20dB, 1ms 0dB, 5s -20dB, 2ms 0dB, 5s -20dB, 5ms 0dB, 5s -20dB ... 500ms 0dB, 5s -20dB.
3. run that pattern through a compressor with various threshold/ratio/soft (if applicable) and/or attack/release settings.

this will tell you how the compressor reacts over time, which gives you a glimpse at its sidechain algorithm. look at the shape of the attack/release phase. is it linear, logarithmic or exponential? or is it sigmoidal? is the release phase the same length for every burst, or does it get longer as the bursts get longer? look for smoothness too. are there jagged edges? in the case of Polysquasher, look at the attack stage and i'm sure you'll find that it starts to compress before the burst. this is the look-ahead, or "preview," action. in some cases, this will be active even when the Compression is supposedly to be off, which could explain your findings. IMO, these are the things that determine the sound of a compressor.

again, i'm not trying to be a dick or get emotional about this. i just try to give some friendly advice when i can.

take 'er easy.

== chunk

[kylen]

Thanks chunk for taking the time to explain compressors to me (like you did last time) and give me a nice burst test case to play with.

It's true I'm looking at a snap shot of a test tone when I made my transfer curve. That's why I was hoping there was a tool or something that would plot that in a real-time window (of a certain time duration). So yeah, a snapshot is a long way away from vibrating molecules over time or complex music - true enough. I've listened to a snapshot (by mistake), a nice 1ms impulse to your brain is not that satisfying!

At any rate I'm hearing a difference in my compressors - the sound is something like pushing down from above (Kjaerhus GPP-1, Polysquasher) sort of a limiting the sound vs oceans of sound organically bubbling up from within the sound space (Voxformer, Tracks VST Compressor). That's my problem - I hate saying stupid stuff like that to describe what the tool does !

Anyway - thanks again for the discussion and your time in explaining details. Now I need to chew on it for a while...till I come up with the next science experiment

[citizenchunk]

Thanks chunk for taking the time to explain compressors to me (like you did last time) and give me a nice burst test case to play with.

no problem. my pleasure.

At any rate I'm hearing a difference in my compressors - the sound is something like pushing down from above (Kjaerhus GPP-1, Polysquasher) sort of a limiting the sound vs oceans of sound organically bubbling up from within the sound space (Voxformer, Tracks VST Compressor). ...

the main differences that i see between these compressors are:
a) GPP-1 and Polysquasher both use a look-ahead (preview), which will grab more (if not all) the transients, whereas Voxformer and VST Comp do not, and will let more transients through. transients are our friends, and make music punchier/dynamic. (i know you probably know this, but some people forget.)
b) the first 2 are probably feed-forward designs, while the second 2 are probably feed-back. please, anyone, correct me if i'm wrong here.

perhaps that sheds some light on the situation.

== chunk