Technical Compressor Question

When the signal goes below the threshold. This is how long it takes before it will release and get prepared for the next attack. If another spike comes in before that time, it will just remain at that level and chop it off, i.e. the attack setting won't come into play. It'll keep squashing until the signal comes down under the threshold for the release time.

This assumes no auto-release, which many compressors have, and that can cause a faster release based on the nature of the signal, which is often desirable since most tracks are not super-consistently either all widely spaced peaks or tightly spaced peaks.

thanks guys. That's what I thought. But I've been reading a book (Mixing Audio) that insists that this is not the case, that the timing stage where release is determined doesn't even know what the thresh is. It explains it well enough to convince me that this is the case - there definitely seems to be situations where the release begins before the signal drops below thresh, in fact it might always be the case that it works this way. The idea is that when the attenuation amount determined by ratio equals the amount allowed by attack then release starts.

The more I think about it, it seems that if the release began when the signal crossed the threshold there'd be nothing to release as the gain reduction as determined by the ratio would already be 0. Release seems to me more and more to be a way to smooth the transition around the threshold, necessarily starting sometime before it and continuing a bit after it.

But I feel like I'm hanging out there somewhere between a very detailed user-level explanation and a precise mathematical one, which I might or might not be able to follow. I was hoping some local gurus would help clarify this. Sorry if this borders on esoterica - I did say it was a technical question.

I've been reading a book (Mixing Audio) that insists that this is not the case

I have this book. Can you refer me to the specific page where this is stated? Maybe you're misinterpreting the text, or maybe the author is just wrong (there were a few technical errors in the book as I recall).

That book, though great for content, really needs the help of a professional book editor. It's crawling with grammatical errors, including the occasional incomplete or nonsense sentence. It has a number of homonym spelling errors, the telltale sign that they relied on a spellchecker alone to clean up the manuscript. So it wouldn't be surprising if someone had a hard time interpreting a certain sentence or paragraph.

I'm pretty sure Bob (bbdude) is correct. The whole purpose of the release parameter is to keep the compressor from being over-responsive to short-term dips in an otherwise hot section. But if we're all wrong about that, I'd sure want to know.

Hi Michael,

In simple the (on edit: I meant "the simple" --- egad!) compressor I wrote, and I believe in most other than certain limiters, the "signal" is not the instantaneous signal, but an "averaged" one where "average" can be quite varied in definition (i.e. subject to significant modification). Also the gain is not changed suddenly, but gradually --- a little at a time. If my compressor did not look ahead, it would be responding to the "signal" a little late. Because it does look ahead, it hopefully reacts at just the right time.

Some of what you posted makes some sense when condsidering these points. I'm concerned that you, as you yourself surmised, are applying comments made by someone who knows how a compressor works to an incomplete understanding of how they work. In my opinion, to understand the comments you would have to have a better understanding of what a compressor is actually doing internally rather than the typical understanding related in music magazines, Internet posts, etc. where a "few details" are usually left out!

Regards,
Dave Clark

For parallel compression, I generally use a limiter. For that application, you generally don't want any attack at all. It should be all body, to be brought up under the regular drums tracks that have all the attack. As long as the release is long enough to supress the initial attack of the drums, that's usually good enough. So something like 10ms should be fine, but you'll have to play with it to taste. Maybe a little more. You want to suppress the attack but let the body through. Since drums have fairly short attack (some are more resonant and some are less, so it's hard to give a fixed answer), you generally don't need a very long release.

Though, this confuses the discussion a bit since limiters always have a zero attack time, and the release is all you get to set. So it attacks immediately upon reaching the threshold, or potentially a little before according to the type of limiter, and then you just need enough release to suppress the attack and perhaps some trailing resonance if the drums aren't very damped.

That works for me anyway.

I was speaking overly broadly there, yes. They can have an attack time. But I was speaking of a brick wall limiter, which I think is what most of us tend to think of when we think limiter vs, a compressor, at least I do, and it's the kind that's appropriate for what I was responding to him about. Actually, I've never had a limiter what wasn't a brick wall limiter, now that I think about it. If it wasn't a brick wall limiter, I'd just a regular compressor, since the point (to me) of a limiter relative to a compressor is to guarantee no signal over the threshold, and you can't do that without a zero attack time, it doesn't seem. I guess there's a reason to allow the attack to go over the threshold then insure that nothing else past that attack time does.

But anyway, yes, you are correct.

ORIGINAL: bitflipper

I've been reading a book (Mixing Audio) that insists that this is not the case

I have this book. Can you refer me to the specific page where this is stated? Maybe you're misinterpreting the text, or maybe the author is just wrong (there were a few technical errors in the book as I recall).

That book, though great for content, really needs the help of a professional book editor. It's crawling with grammatical errors, including the occasional incomplete or nonsense sentence. It has a number of homonym spelling errors, the telltale sign that they relied on a spellchecker alone to clean up the manuscript. So it wouldn't be surprising if someone had a hard time interpreting a certain sentence or paragraph.

I'm pretty sure Bob (bbdude) is correct. The whole purpose of the release parameter is to keep the compressor from being over-responsive to short-term dips in an otherwise hot section. But if we're all wrong about that, I'd sure want to know.

Hi flipper, I agree with you, that book certainly could have used some editorial review, though I appreciate the length to which the author goes to try to explain some things. There are quite a few places that have required some interpretation, but this is first place I just got bogged down - though I did have a talk with some EE guys last night that helped me a bit and convinced me that Roey is right, just not clear enough in the explanation for me.

Anyway, the dicussion of this begins on page 284 and is illustrated in fig 16.12. There is even a highlighted 'bullet' on the page that states "As opposed to what many sources state, the release function is not related to the signal dropping below the threshold".

My conclusion at present is that attack and release serve to parameterize two curve smoothing functions (aka 'timing laws') used to modulate the gain reduction of the signal around the threshold, giving us the ability to do just what you said - tame the compressor a little in very dynamic sections. I think my original question as to what triggers the beginning of the release period is being replaced by what ends it?

Thanks everybody who has commented here. Again, sorry if I'm digging into unecessary detail - I'm not sure the answers I seek will change how I (or you) use the compressor - it's just that I've got this inquiring mind and it wants to know. I cannot do much to tame it after the scotch runs out. It has gotten me in lots of trouble already, since I was a child. 'Why?' is not a welcome question all that often in the world. So I'm just resigned to my fate. thanks for your indulgence here....

There is no such thing as "unnecessary detail", IMO. I love to go deep on this stuff. But then I'm an unapologetic gearhead.

Mr. Izhaki does seem to be quite clear on this, explaining that release begins immediately after the end of the attack period, like an ADSR envelope generator on a primitive synthesizer without the D or the S.

I'm going to have to think on this a bit...

EDIT:
Thanks, altima_boy, that paints a clear picture.

The key is that compression is re-enabled with every qualifying sample that occurs after the attack period has completed. A consistently high series of samples would keep the compression happening and be continually resetting the release counter. Compression would therefore continue to be applied as long as the samples stayed above the threshhold, regardless of the release time.

So the release phase is not triggered by falling below the threshhold, but rather is prevented from starting until we fall below the threshhold. A fine distinction, and the ultimate effect is the same - a smoothing of the effect.

The way I've read it explained, many times, though of course I could have been misunderstanding is:

1. The signal starts coming up. As it hits the threshold (or approaches if a soft knee), the compressor clamps down
2. It keeps clamping down, watching for the signal to fall below the threshold, unless of course it's an auto release type compressor, which so many are, in which case it automatically releases at some point perhaps before the release period really ended.
3. Once the signal falls back down to the threshold, the release time starts and it won't reset until the release time ends.
4. Until it releases, it's still clamping down, so another attack that comes in before it releases will be clamped down on from the get-go. It'll still come up above the threshold since it'll be a high peak probably and the amount of clamping isn't enough to stop it, but it'll have less attack since the compressor is already compressing, and the release counter starts again at this point.
5. Once the manual release time or the auto-release time is reached, it'll reset, and start looking for another attack, and it'll let that attack through up to the attack time, and so the cycle starts over.

That's always been my understanding. Is that not correct? If the release period was unconditional, and started at the attack point, then they would release at the end of the release period, even if the signal was still above the threshold, and that doesn't seem at all what is observed in reality. They keep clamping down until the signal gets down under the threshold.

If the purpose of the release time is to prevent it from flailing about if the signal goes up and down around the threshold, then it would seem it would only be useful if it started once the signal fell below the threshold, like the Hold of a gate. If the purpose is to just prevent the compressor from attacking and releasing too fast, then I guess it would make more sense if it was started from the start of the attack kicking in.