Parallel Compression – what’s really going on?

[Ostermilk]

Rather than a listening test a better test for this would be for someone to provide a parallel compressed track and for you to obtain a null* from it using the same compressor inline. Otherwise you know what'll happen you'll get the inevitable 'this one has more 3 dimensional warmth somehow than the other') type comments which don't demonstrate anything one way or another.

*Reason's limitations in being able to produce a perfect null notwithstanding.

I've done it before by the way in another DAW with a stock compressor, and yes if might be surprising to some how close it can be and call into question whether it's even worth doing as opposed to just using an inline compressor, two things are interesting though, one is the remaindered signal even when you've got close to producing a null (it normally contains some good punchy stuff) the second is that when you alter the relative levels in the parallel path is where you start to notice that all bets are off.

So two thing are important in achieving a worthwhile effect with parallel compression (aside from having an extremely low threshold and a ratio set to limit or close to) is to apply worthwhile amounts of make up gain to the compressed signal and the other is to alter the relative levels between the two signals to taste.

I gave an example earlier where I would (and do) use it and that is where I've got a pre-mixed acoustic drum loop and I want to make the kick, snares and toms to pump and breathe like a beast, but a compressor at this setting would normally wipe out the cymbal swishes and any distinct hi-hat patterns too during the slow release I'd want for this effect, but by reducing the relative level of the compressed signal you are often able to get both things happening nicely.

If you had quoted just a tiny bit more of what I said you would have answered your own question.

The line before what you quoted said this:
"So no, while you can get extremely close to canceling with gains at different levels, it's not going to cancel at any setting I've found like it does when they are equal level."

So I just specifically said you cannot null at any other setting than equal levels, and yet you ask me to try to null a file I just said cannot be nulled.

I'm not sure why you keep on with this approach here…

You are creating a straw man argument, suggesting I've said something I have not (that you can null at any parallel level) and then arguing against that point (saying you can't, which is EXACTLY what I've already said).

And it sounds like you're trying to 'sell' me on the idea of parallel compression, something I've been using since 1984. I "get" it. I use it.

Unless you've got something more to add to this conversation, we're done here OK?

I wasn't selectively quoting to invalidate anything you said, nor was I asking a quetion.

I simply quoted that part with the suggestion that a listening test probably wasn't the best idea in the circumstances and gave my reasons why.

So what is this approach I'm supposed to be keeping up here?

There's some good suggestions there to help your readers to get the most out of using parallel compression, no?

Come round for a beer fella, honestly I'll bet we've got more in common than we have that divides us. Seriously.

[selig]

If you had quoted just a tiny bit more of what I said you would have answered your own question.

The line before what you quoted said this:
"So no, while you can get extremely close to canceling with gains at different levels, it's not going to cancel at any setting I've found like it does when they are equal level."

So I just specifically said you cannot null at any other setting than equal levels, and yet you ask me to try to null a file I just said cannot be nulled.

I'm not sure why you keep on with this approach here…

You are creating a straw man argument, suggesting I've said something I have not (that you can null at any parallel level) and then arguing against that point (saying you can't, which is EXACTLY what I've already said).

And it sounds like you're trying to 'sell' me on the idea of parallel compression, something I've been using since 1984. I "get" it. I use it.

Unless you've got something more to add to this conversation, we're done here OK?

I wasn't selectively quoting to invalidate anything you said, nor was I asking a quetion.

I simply quoted that part with the suggestion that a listening test probably wasn't the best idea in the circumstances and gave my reasons why.

So what is this approach I'm supposed to be keeping up here?

There's some good suggestions there to help your readers to get the most out of using parallel compression, no?

Come round for a beer fella, honestly I'll bet we've got more in common than we have that divides us. Seriously.

Was just going to say the same thing - which is why I keep struggling to understand why you keep disagreeing with me!
I keep pointing out that I have PREVIOUSLY made the statements that agree with what you are saying!

Specifically (in the last post): You suggested someone "provide a parallel compressed track and for you to obtain a null* from it using the same compressor inline" when I has JUST pointed out it WON'T null. I was reminding you that in the line you DIDN'T quote, I already pointed out it would not null, that's all.

I really don't want to put too much into tests that WON'T null since all they prove is that there is "some" difference". That difference could be timing, could be level, or it could be an actual sonic difference - we just don't know.
BUT, we DO know that a test that nulls proves there is no difference, and by extension a test that nulls the same or more as the bypassed version (the "control" in this case) therefore passes the same test.

Your friend, the great all-knowing Selig aka Captain Obvious, someone who ain't going take a blind bit of notice of any evidence put in front of them, and someone who truly enjoys the fragrance of their brain farts not to mention fruitless diversions for the sake of conversation.

[Ostermilk]

Your friend, the great all-knowing Selig aka Captain Obvious, someone who ain't going take a blind bit of notice of any evidence put in front of them, and someone who truly enjoys the fragrance of their brain farts not to mention fruitless diversions for the sake of conversation.

There's nothing wrong with any of those things in my book, they're all things that seem to be manifest in all of us when we try and peel off the chaff in an attempt to reveal the pearl of good stuff that lies beneath. Someone once said that an argument will get more heated in direct proportion to how little is at stake... There's a lot of truth in that.

To be clear, it's never my intent to fall-out with anyone here and trust me you'd be very low on the list of people I would want to bug even if that intent were to change....

[EdGrip]

Oh yea, your question about resolution with Selig Gain - you can't get more resolution by using two of the same resolution devices together.

Yeah, I realised this on thinking about it again this morning. I feel like my brain was on to something, though - how about:

Split <
A(Gain1)
in parallel with
B(Gain2>Gain3)
> Merge.

So you set your signal not far off with A, and then with Gain2 set low, you can use Gain3 as a fine-tune? Assuming there's no latency.

[scifunk]

I thought parallel compression meant mixing a heavily compressed single in to a non compressed original one to provide some fatness to the original signal while maintaining the dynamics/higher frequencies.

[EdGrip]

I thought parallel compression meant mixing a heavily compressed single in to a non compressed original one to provide some fatness to the original signal while maintaining the dynamics/higher frequencies.

That's what this whole thread is about. Read it, it's interesting.
Selig proposed that parallel compression gives results that are so similar to just changing the compression ratio that they're functionally the same, or at least not as different as The Internet would have you believe.

[Marco Raaphorst]

This difference imo are dynamics. With a compressor using a sine wave you can match volume compression. But on dynamic material you get totally different results when blending the original channel with the compressed signal.

[selig]

This difference imo are dynamics. With a compressor using a sine wave you can match volume compression. But on dynamic material you get totally different results when blending the original channel with the compressed signal.

I didn’t use a sine wave, I used dynamic material.

BTW, this was a proof of concept piece designed to get a conversation going. Which it seems it did!


Sent from some crappy device using Tapatalk

[Marco Raaphorst]

This difference imo are dynamics. With a compressor using a sine wave you can match volume compression. But on dynamic material you get totally different results when blending the original channel with the compressed signal.

I didn’t use a sine wave, I used dynamic material.

BTW, this was a proof of concept piece designed to get a conversation going. Which it seems it did!


Sent from some crappy device using Tapatalk

We can only hear it, not explain it, it seems. Perfectly fine with me. My technical knowledge is rather limited.

[normen]

We can only hear it, not explain it, it seems. Perfectly fine with me. My technical knowledge is rather limited.

Well it depends on what you do. If you have for example a compressor like the famous Neve 33609 then that particular compressor will create most harmonics when its reducing by about 6dB. So if you want those harmonics but not 6dB of gain reduction (i.e. a lower ratio) then you will have to parallel compress. You will also get only a certain amount of that character though

For a totally clean digital compressor it really doesn't make much sense.

[selig]

I thought parallel compression meant mixing a heavily compressed single in to a non compressed original one to provide some fatness to the original signal while maintaining the dynamics/higher frequencies.

A lower ratio will accomplish the same (or similar) thing in many cases. When you use parallel channels you increase the gain by 6 dB. So to restore the original gain you reduce the results by 6 dB - meaning the original signal is now 6 dB lower than before.

All to say there is little point in using parallel compression unless you also make other changes as well, such as adjusting the balance between the two channels to a significant degree, or adding saturation/EQ/filtering to only one of the channels in addition to compression.

IMO, there would be little point to compress in most cases and expect to maintain the original dynamics - that’s the point of compression, right, to alter dynamics?


Sent from some crappy device using Tapatalk

[selig]

I thought parallel compression meant mixing a heavily compressed single in to a non compressed original one to provide some fatness to the original signal while maintaining the dynamics/higher frequencies.

That's what this whole thread is about. Read it, it's interesting.
Selig proposed that parallel compression gives results that are so similar to just changing the compression ratio that they're functionally the same, or at least not as different as The Internet would have you believe.

There ARE differences…
As others pointed out, the differences come when you make other changes, such as a significant change to the balance between the wet and dry channels (as opposed to keeping them equal level).


Sent from some crappy device using Tapatalk

[WeLoveYouToo]

yeah, a dry track mixed with a compressed track will simply give you a half compressed track with no more thickness (or whatever effect parallell compression is supposed to give) than before.
the oscillation of the dry track will stack normally on the uncompressed, parts of the audio, and proportionally less so on the transients etc depending on how much compression is applied... and so you are left with exactly a half compressed track.
for any sort of sonic advantage to using two of the same parallell tracks some sort of distortion would need to be applied in the compressor on one track and not the other, which a physical analog compressor would do to a degree, or adding another effect in the chain on the wet side.

[raveled]

And now there are developers that skip the ratio control altogether (which has to be matched with threshold and make up gain), reasoning that a mix control with automatic gain compensation is equivalent and much easier to use. Must have read Selig’s article.

https://www.newfangledaudio.com/post/oo ... d-mix-glue

[selig]

And now there are developers that skip the ratio control altogether (which has to be matched with threshold and make up gain), reasoning that a mix control with automatic gain compensation is equivalent and much easier to use. Must have read Selig’s article.

https://www.newfangledaudio.com/post/oo ... d-mix-glue

Ha! I did that with Selig Leveler over 10 years ago - maybe should have made a bigger deal over it? That was where this strange and wonderful journey began, when I realized the Leveler’s Balance control on early prototypes was acting exactly like a ratio control…I just came at it from the other side of the equation!

Since this was originally written I’ve done further tests that only strengthen the idea that parallel compression on it’s own is just a ratio control. You have to adjust other controls to get better nulls, but there are certain relationships such as Inf:1 parallel = 2:1 insert compression (as previously mentioned) but also other ratios can also be fairly closely matched (*see below) with adjustments to overall balances and thresholds.

(Wow, in re-reading this thread I forgot how “passionate” folks can get over technical subjects, and especially without posting any technical examples of their own to refute my findings! In re-visiting this subject I feel even stronger about what I originally wrote, but as always will remind folks that all the original “exceptions” still apply: this only works with fairly simple compressors (no “auto” features, no saturation, etc) and no other processing in the chain such as saturation or filtering/EQ. Additional processing beyond just compression is still where the advantages of parallel processing shine IMO though I haven’t used/needed parallel processing myself in years now.)

*And I’ve covered this in my monthly Recording Magazine column (Field Notes) recently, but a “close” null test is actually MUCH closer than I used to think. So in those cases where I’m only reducing the signal by 60-70dB and not getting a “perfect” null, here’s a bit of a reality check. If ALL you do is invert the polarity of a duplicate audio channel and reduce the gain by 0.1dB, you’ll only get around a 45dB “null”. Which means any better/lower null result means there is even less than 0.1dB difference! For every 1/10 reduction you get another 20 dB less residual with a null: 1dB difference = -25dB residual, 0.1dB = 45dB residual, 0.01dB difference = -65dB residual, continuing all the way down to 0.00001dB leaving-125dB!
And it’s even worse with timing differences - even just one sample delay on white noise gives you less than a 3dB “null”, or hardly any cancellation at all (about the same as any two non-correlated signals when summed).
All to say, a null leaving 70dB residual audio is pretty damn close!!!

[The more you know: to those curious as to how I’m measuring down to 0.00001dB difference, I ‘m using Selig Gain, of course. When you automate the gain values you can type in values that low, which is 1 x 10^-5 (scientific notation) or one hundred-thousandth of a decibel.]

[raveled]

Sorry, of course I should have written “And now there are more developers…”

Thanks for interesting and somewhat nerdy elaborations