Parallel compression under microscope

[Vitaly]

Hello,

Recently, I was investigating Parallel Compression for myself and I've come to unexpected results. In fact, I've come to the conclusion that there is no any real advantage Parallel Compression has over Downward one. The only thing Parallel Compression gives is the distortion of the shape of Attack, Release time. Also, it affects the shape of compression ratio. The null test shows -45...-30dB difference (depends on music) between Parallel Compression and equivalent Downward compression. I put all my thoughts in an article below. I thought you might be interested to have a look.

Parallel compression under microscope

Parallel Compression Calculator

Any comments are welcome.

Regards,
Vitaly.

[Lagerfeldt]

While your procedures and resulting findings may lead you to draw those conclusions, I think it's safe to say that they are your conclusions only.

I like to use an extremely fast attack and a above medium ratio, go well into the signal (e.g. -40 dB) and then control a lot of the sound using the release time.

Bring it back to the mix for a fuller sound and lift the low passages, very effective!

Check out this picture for an example of how parallel compression can add fullness and lift low passages without changing the transients.

I.e. notice the low passage has been raised and generally a fuller sound has been obtained - but the transients (peaks) are left mostly untouched, unlike downward compression.

So in conclusion there certainly are benefits to this way of compressing, which many both mix and mastering engineers use daily with great success.

Ironically I'm writing an article for Scandinavia's biggest musician's magazine today about parallel compression, and not surprisingly my recommendations will be different from yours ;-)

[Adam Dempsey]

It has to depend on the purpose/sonic intent: to reduce dynamic range from the top down or the bottom up.

In any case, subtlety is most often the key. The parallel path is often mixed way below the direct path with, for example, just 2 - 2.5:1 ratio, real fast attack (prevents overshoot), ie: the hotter the signal (above threshold) the less compressed signal is in the final output. And yes the resultant output's transfer curve looks like a soft knee, variable ratio - in fact just like the "old" Sony DAL-1000.

[Nordenstam]

Originally Posted by Vitaly

Also, it affects the shape of compression ratio.

And there's the crux of the matter.

SIDE CHAIN COMPRESSION

[turtlerock]

if you want that article published in audio technolgy in our part of the world pm ( aust/NZ) me and i will join the dots for you.

maybe if was printed on paper i would have some hope of understanding it ...!
thats a lot of heavy looking graphs i am amazed people think like this !
but i am amazed at a lot of things this month

me...? i plugged a sontec limiter into the reverb send/return on the neumann cutting console and invented side chain compression for myself 20 years ago
because i could not get the sontec to sound any good on its own
and i have found SC to be very usefull over the years

i was bummed when i found out it was not my invention !

perhaps i should have made up a new age sounding name for it and wrote a book about the secret process i developed 30 years after it was first used

i am serious about publishing thing though

[heathen]

Originally Posted by Vitaly

Hello,

Recently, I was investigating Parallel Compression for myself and I've come to unexpected results. In fact, I've come to the conclusion that there is no any real advantage Parallel Compression has over Downward one. The only thing Parallel Compression gives is the distortion of the shape of Attack, Release time. Also, it affects the shape of compression ratio. The null test shows -45...-30dB difference (depends on music) between Parallel Compression and equivalent Downward compression. I put all my thoughts in an article below. I thought you might be interested to have a look.

Parallel compression under microscope

Parallel Compression Calculator

Any comments are welcome.

Regards,
Vitaly.

What happens when you add an eq into the equation? Which I do probably say 99.9% of the time when using parallel tracks. I may use 3 parallel tracks sometimes if I'm not getting the goods from 2.

[Vitaly]

Originally Posted by Lagerfeldt

Check out this picture for an example of how parallel compression can add fullness and lift low passages without changing the transients.

That is what we all think. What I'm saying is that you can get the same result by Downward compression (moreover, it will sound even fuller). You just need proper settings. The reason why you got the transients unchanged isn't Parallel compression. It is a short Attack time you set. I don't want to argue with you just try that calculator. Keep Attack and Release the same for both Parallel and Downward and recalculate settings you use in Parallel compression into the settings for Downward one. Then compare the sound. It can save your article

Regards,
Vitaly.

[Lagerfeldt]

LOL, I'm sorry but that's just a bit silly.

[Adam!]

I've been lurking gs for years now, but this was the thread that finally made me register. Vitaly, I could find no fault with your math, but I figured that you had underestimated the importance of the attack and release times. Either that or, under fairly extreme gain reduction, some subtle flaw might come to light. So I gave this a try, setting a parallel compressor to crush and blending it in very loud, and then setting up a regular compressor using your calculator's recommended settings. To my amazement, they nulled out! I tried different settings, long attacks, short releases, the works, and they all came well within a tenth of a db of nulling out (close enough that I could not tell them appart in a blind test).

So, you've convinced me: parallel compression--while seeming intuitively different--is actually just a more complex way to do regular ol' plain-jane downward compression. As Heathen says, this finding doesn't apply when your eq'ing the parallel buss (although I suspect that one might achieve an identical result to this using multiband compression... perhaps you could make that calculator next!), or in mixing applications (like, say, using parallel compression on just the kick and snare bussed together, etc), but as far as mastering goes this is a very important observation.

[Darius van H]

I like the fact that some plugin compressors have a mix control - that's parallel compression -

so, if you think, "hmmm, that's great, but i'd like to make it a bit more subtle" you can yank the mix control.

Other then that i find parallel compression an overly complicated way of doing something, and i agree with the above poster that there's nothing intrinsically different about it from normal comp.

[kjg]

Originally Posted by Vitaly

That is what we all think. What I'm saying is that you can get the same result by Downward compression (moreover, it will sound even fuller). You just need proper settings. The reason why you got the transients unchanged isn't Parallel compression.

Dear Vitaly,

I appreciate the efforts you made investigating this matter, writing about it, and making the calculator.
It is true that you can get close using the right settings. It sounds similar. But not the same. The one thing that doesn't get very close at all is the sound of the transients. Which is exactly why mastering engineers like parallel compression on mixes, and mixing engineers like it on e.g. drum buses.

Let's investigate this classic drum buss example, shall we?

I've taken a drum loop, set up parallel compression for a sound I might hypothetically use in a mix, and then used your calculator to set up ordinary downward compression to resemble it.

The compressor used is the Waves Rcomp (I tried the PSP Mastercomp, too - same results).
These are the settings shared between both compressors: attack 10, release 100, threshold -20, manual release, opto, warm (I tried it with the shortest attack time too - same results).
The parallel comp uses a ratio of 9 to 1. The downward comp uses a ratio of 1.33 to 1.
The downward comp is compensated in gain by 6.02 dB.

And yes, it sounds similar, but no, is does not sound the same. The attacks in the downward comped signal are squashed. The attacks in the parallel comped signal are more alive. This is exactly why we use parallel comp. Hear for yourself - I think Lagerfeldt's article will be fine.

These samples are truncated to 16 bit after applying GPDF (gaussian) dither without noise shaping and compressed with the latest Lame mp3 compressor (320 kbps, stereo, all filtering disabled, optimized for quality).

Even in mp3 format the differences are striking. It is not like we are talking about some tiny, quasi-placebo stuff going on at -140dB, which can only be heard with 20k speakers and golden ears! Listen to the difference in the snare drum and hat. The null test file is not even close. There is too much in there to even speak of transients - it's complete attacks of the drums.

Again, I appreciate your effort - I hope you can appreciate mine too..

Finally, let me try to answer the questions you posed in the conclusion of your article:

"What is parallel compression? A method allowing us to achieve a 'different' sound or another way around?"
Not another way around, a different sound. Not 'different'. Different.

"Is it worth setting up busses, auxiliaries, sends, and returns, ..."
Yes, obviously.

"..., and then thinking about how much you should add?"
No you don't think about how much you should add. Your calculator does not tell you how to set your eq either, does it? You balance the dry and wet faders (or the convenient mix control Darius likes to yank around) until you like how it sounds.

"Or, is it much clearer to go the standard way?"
This is like asking if it is easier to stand in the rain than to take a shower. Yes, both get you wet... But it isn't the same thing, is it?

With kind regards,
Klaas-Jan Govaart

[Bob Olhsson]

To me, parallel compression does two things. It reduces the compression ratio and it hides the sudden onset of distortion when a peak hits the threshold although the down-side is that material will be more distorted overall.

Sometimes it enhances detail in a mix while others it mucks things up. Like everything else in mastering, it's all about monitoring and making sure the mix really sounds better rather than just a bit louder but otherwise worse.

[Vitaly]

Dear Klaas-Jan Govaart,

Thank you very much for your effort. You have got the same result as I have just let me explain something. The null test will never give us... null. Why? Because when we mix the transfer curve of Dry signal, which is linear, and the transfer curve of a compressor, which is logarithmic, we get some kind of 'hybrid' curve. The null test shows us the difference between that 'hybrid' and logarithmic curves. The same thing happens to Attack and Release envelopes. The time stays the same but the envelope becomes changed (the hybrid of the linear and exponential curves). It is these distortions we can see in the null test. If you turn down the track with the downward compression (-0.5 dB maybe) the null test will give you even smaller difference. Depends on music it is circa -45...-35dB. The transients stay exactly the same and here they are. Look at your snare attack:


Yes, the chain EQ+Comp is more complicated and yet I'm 100% convinced that we won't find any advantage Parallel compression has over Downward if we recalculate settings carefully.

As I said, I don't want to fall into I-can-hear-the-difference arguments. There is a very small difference. Whether it is worth setting up the parallel scheme is totally up to you.

Adam! You wouldn't believe what I felt when I came to this

Kind Regards,
Vitaly.

[DominicWyeth]

I like compression that sounds nice


.. I also like turtles

[Lagerfeldt]

Originally Posted by Vitaly

Whether it is worth setting up the parallel scheme is totally up to you.

Many compressors have a mix setting, allowing for direct parallel compression without the need to setup busses, etc. Very easy.

Again, further processing of the parallel signal (e.g. with EQ) is an option not available using regular compression.

So parallel certainly has its benefits in this area too.

[Nordenstam]

G'day! =)

Originally Posted by Vitaly

The null test will never give us... null. Why? Because when we mix the transfer curve of Dry signal, which is linear, and the transfer curve of a compressor, which is logarithmic, we get some kind of 'hybrid' curve. The null test shows us the difference between that 'hybrid' and logarithmic curves.

Are you saying there is no difference in the intrinsic operation of the compression schemes, but in practice there is a difference that only nulls to -50 at best? Seems like a difference to me..

I'm not sure about the hybrid curve explanation. Still seems like the matter is as the text book says: pushing loud parts down or soft parts up. Please help me understand this if that's not the case.

Did a quick experiment with this. Looking at the crest ratio (difference between RMS and peak levels) shows that the compression options indeed does what the theory says. Downwards compression gives smaller crest ratio in the loud parts, while parallell compression gives smaller crest ratio in the soft parts. To my understanding, that's exactly why people use parallel compression.

I used the sound forge compressor as that was handy and had all the options needed available in the interface. The settings used where threshold at -35, ratio 22:1 for parallel - corresponding to downards with 1.2:1 ratio and 6.02dB gain. Same attack/release settings used throughout. Both tracks had huge dynamic range from very soft to loud.

Drum track: 
Downward loud part crest ratio: 19.140dB
Parallel loud part crest ratio: 19.761dB

Downward soft part crest ratio: 28.698dB
Parallel soft part crest ratio: 28.501dB


Complex music: 
Downward loud part crest ratio: 14.458dB
Parallel loud part crest ratio: 15.189dB

Downward soft part crest ratio: 23.734dB
Parallel soft part crest ratio: 23.531dB

Both tracks gave about the same result: 0.621dB and 0.731dB difference in the loud parts, 0.197dB and 0.203dB in the soft parts.


Regards,

Andreas Nordenstam

[PatrikT]

Originally Posted by Vitaly

In fact, I've come to the conclusion that there is no any real advantage Parallel Compression has over Downward one.

Congratulations!

Scientific BS apart - it can kind of be heard too...


BRGRDS
Patrik

[kjg]

Originally Posted by Vitaly

You have got the same result as I have just let me explain something.

snip...

Explanation won't be necessary, thank you. We both understand what is going on. You are just drawing the wrong conclusions.

Of course you are excited about your "discovery". But it is not really a discovery - it is an insight. You have realized that two techniques are very similar. Like many, many others before you. Still, good for you!

You are then presenting numbers that almost match up. The differences are in your opinion mere distortions and of minor importance ('just' -40dB - so the theory is still valid...).
You seem to reason somewhat like this: "I've come to realize that the differences between downward and parallel are not as big as I first thought! I actually find them very small! Therefore, I conclude that the more "complicated" technique doesn't really have a use anymore."

This is where your reasoning is off. These slight differences are the main reason the alternative, "complicated" way was in use all along. There was never anything else! I'm sorry that you were expecting a more spectacular difference and are now a bit disenchanted. But don't present your newly found insight like you discovered a new continent, please.

From your article: "I advise you to stop reading and start experimenting, because the conclusion you are about to come to will change your attitude about parallel compression forever!"

Right...
I'm happy for you you learned so much in the process of experimenting and writing, but I'm sure that for a lot of people here their attitude towards parallel compression hasn't changed at all, because:

a) It is still the same technique, with the same differences from downward compression.
b) These "small" differences are the exact reason why they used it in the first place.

Just because you have now realized what is actually going on, doesn't mean it changes much for others. They already knew. And they were making their informed decision for the one technique over the other just as easily without you knowing...

"if a tree falls in the forest and nobody is there to hear it, does it make noise?"

Just because you were not there in the forest to hear the tree fall, doesn't mean it didn't make a noise. Others were there. They already knew the difference was small, and still used it. It is small to you, but it is still a very real difference.

Originally Posted by Vitaly

Yes, the chain EQ+Comp is more complicated and yet I'm 100% convinced that we won't find any advantage Parallel compression has over Downward if we recalculate settings carefully.

This is just silly. It is different. You can hear that advantage. You are presenting pictures to proof that the results of the techniques are the same??? I'm sure you can't see the difference between different dithers either. Are you saying they can not be heard too? Get real!

Originally Posted by Vitaly

As I said, I don't want to fall into I-can-hear-the-difference arguments. There is a very small difference.

Yes, there is. And this difference is one of the reasons (not some bigger difference you were expecting to find) why the technique is used. It is not just a minor flaw in your 'theory', but one of the actual reasons!

Lupo brings another difference to our attention. Another reason why parallel is so useful for mastering.

Originally Posted by Lupo

Downwards compression gives smaller crest ratio in the loud parts, while parallell compression gives smaller crest ratio in the soft parts. To my understanding, that's exactly why people use parallel compression.

In parallel compression, when the signal gets louder, the relative contribution of the compressed path to the sum gets smaller.
At the same time it increases intelligibility of quieter passages, because there the contribution of the compressed signal to the total signal is relatively large.

That is more similar to upward compression, isn't it?

A downward compressor can sound quite similar to a parallel compression setup. But it doesn't effect the transients/attacks in the same way (don't look at pretty pictures - listen!), AND it doesn't respond in the same way to larger term variations in signal level (e.g. verse/chorus, crescendo).

So, you found an "equivalent" dynamics processing technique to parallel compression. It just doesn't respond the same way to either micro- or macrodynamics. But besides that, it is pretty much the same...

Even when you don't care for the difference in the transients, and are working on material without any macrodynamics, a very practical reason to still use parallel is that a lot of compressors don't offer the possibility to set very low ratios accurately. Take for example the API2500. Try dialing in 1.33, 1.25, 1.2, 1.17, 1.15 or 1.1 to 1 on that. With many compressors, there is no other way to achieve these kind of ratios.

Still, of course it is all irrelevant to someone who can't hear the difference between two signals nulling out to -40dB, or a difference of 0.7 and 0.2 dB respectively in crest factor.

Let's see some more pictures!

regards,
kjg

[heathen]

Originally Posted by Bob Olhsson

To me, parallel compression does two things. It reduces the compression ratio and it hides the sudden onset of distortion when a peak hits the threshold although the down-side is that material will be more distorted overall.

Sometimes it enhances detail in a mix while others it mucks things up. Like everything else in mastering, it's all about monitoring and making sure the mix really sounds better rather than just a bit louder but otherwise worse.

I agree with Bob here, though from a mix perspective, say a drum group is less than chunky enough within a mix, at some points the drums may sound just right in say a very dynamic piece of music where at other times they are losing a bit of impact, I think this where parallel comp excels in a mix, by getting a good blend of the 2 can help a dynamic and dramatic mix from either becoming to weak in some points or too powerful in others, even automating the 2 between passages can be of great advantage.

To me it may increase apparent loudness in that group but not too much if it's used conservatively, to me it helps keep a stability within the mix. Can help keep some weird transients from popping thier heads out, or make some worse.

[narcoman]

Originally Posted by Vitaly

Hello,

Recently, I was investigating Parallel Compression for myself and I've come to unexpected results. In fact, I've come to the conclusion that there is no any real advantage Parallel Compression has over Downward one. The only thing Parallel Compression gives is the distortion of the shape of Attack, Release time. Also, it affects the shape of compression ratio. The null test shows -45...-30dB difference (depends on music) between Parallel Compression and equivalent Downward compression. I put all my thoughts in an article below. I thought you might be interested to have a look.

Parallel compression under microscope

Parallel Compression Calculator

Any comments are welcome.

Regards,
Vitaly.

Well - your article is quite good - but you've drawn subjective conclusions. They arent infallible. The "small" differences you infer are in fact \HUGE differences. We're dealing with "punch" and transient clarity - these ARE subtle things! But they make all the difference in the world! Good article - but I'd rethink your conclusions. A 40db null is not a reason to discard a technique.

[mikeroephonics]

I recently mastered a punk rock album using parallel compression on all of the tracks to one degree or another. The band and I both love the sound we got. It's great to get a punchy yet dynamic sound with nice transients!

[Nordenstam]

Hey Vitaly, the calculator is cool! Would never have gotten around to making the pic below without it. Thanks!


Could write an article on the subject but there's work to be done.. The pic says more than a similar bunch of KB's in ascii text.




(the axes are linear scale, level upwards and time horisontal. made in soundforge's "graphic fade" display using 10 minutes of pink noise with linear level fade. blue line is dry response. grittyness on the curve is caused by the random nature of pink noise. settings on compressor was threshold -6(50%), attack 0, release 500, ratio 1.3:1 and 2:1 respectively. hope this is a correct way to view level response.. )


Regards,

Andreas Nordenstam

[bmanic]

I think perhaps Vitaly doesn't have a precise enough monitoring environment to actually hear the differences? When somebody says "it nulls to -30 or -40dB" and call it near identical to the original it might perhaps be time to invest in better monitoring? You really need to get nulls down below -90dB or more to call it that, in my humble opinion.

Cheers!
bManic

[Lagerfeldt]

Right.

[Vitaly]

bmanic, where I called it 'near identical'? I said: 'There is a very small difference. Whether it is worth setting up the parallel scheme is totally up to you.'

BTW I'm happy with my Genelec. Thanks.

Guys, apparently you've got it wrong. Nobody is going to take Parallel compression from you My goal was to share the results I've come to as I thought it could be interesting to someone else. There is no need for anger here.

Kind Regards,
Vitaly.

[Lagerfeldt]

I think your work is admirable, but your conclusions should be left out as they are not correct or at the best extremely subjective.

[kjg]

Originally Posted by Lagerfeldt

I think your work is admirable, but your conclusions should be left out as they are not correct or at the best extremely subjective.

Exactly. Nobody is angry here, just not agreeing with your conclusions.

[heathen]

Originally Posted by Vitaly

bmanic, where I called it 'near identical'? I said: 'There is a very small difference. Whether it is worth setting up the parallel scheme is totally up to you.'

BTW I'm happy with my Genelec. Thanks.

Guys, apparently you've got it wrong. Nobody is going to take Parallel compression from you My goal was to share the results I've come to as I thought it could be interesting to someone else. There is no need for anger here.

Kind Regards,
Vitaly.

I thought it was quite interesting, making observations and coming to a conclusion is very good work, but I just feel there is a bit more to it.

Thanks for sharing.

[jefferson]

i see two issues:

1. when doing parallel compression the phase of the original signal is unaltered while the compressed signal is altered, and when using only downward compression the phase is altered in an entirely different way.

2. you wrote your calculator for windows. you should know that everyone here uses macs.

[Lagerfeldt]

2. In mastering PCs are actually very widespread, but otherwise I agree.