Surround compression detector handling

[synthpark]

Hi

For stereo one would usually take the maximum of left and right channel OR sum them (mid signal) to get a common detector signal, or doing mid/side compression.

But what about surround when doing a master compression?

I am not sure whether the same approach works here, probably not. Left/Right would likely be treated like in stereo or also including the center channel, what about the the bass channel? Also, perhaps the speakers left and right behind are compressed separately ...

Any info? Thanks

[BertKoor]

Surround usually means the target is (home) cinema. Luckily there the target is around -25 dB LUFS. (some sources say -23dB, others say -27dB LUFS)
That means compression / limiting is hardly needed, if any at all. There's plenty headroom available for things to suddenly go very loud.

And yeah, you don't want that spaceship passing by on the rear left to duck the center channel. Do whatever feels natural and sounds good. You'll figure it out

[hugoderwolf]

Record companies are massively pushing multichannel/spatial audio for music as well. And anyway, compression may be useful even if there's infinite headroom, simply to adjust dynamic range for the sake of artistic intention.

For a multichannel compressor imho it makes sense to make the detection loosely based on the way the LUFS standard combines channels. IIRC it sums the squared magnitudes of the channels with some weighting factors (depending on channel layout, but it's mostly something like -1.5dB on rears or something), then does the remaining calculations on the result. The approach could be adopted for compression.

[synthpark]

Yeah, I was thinking about some weighting as well, but not sure which channels to weight how. Thanks.

[mystran]

For stereo (or surround) link, if you have a peak sensing compressor (or limiter) then go ahead and take the maximum... but if you have a RMS (or "loudness" which is weighted RMS) sensing compressor, then you probably want to compute the combined loudness by taking the vector norm (root of sum of squares, sqrt(L*L+R*R+C*C+...)) instead, because this estimates the combined power (as an approximation for loudness).

Combining a peak sensing compressor with a vector norm over channels, or an RMS sensing detector with a maximum channel makes a lot less sense, because you're basically treating the envelope and the channel link in different ways. There's no rule against doing it... but like you probably get better results if the detector and the channel linking scheme are doing the same thing.

[synthpark]

For stereo (or surround) link, if you have a peak sensing compressor (or limiter) then go ahead and take the maximum... but if you have a RMS (or "loudness" which is weighted RMS) sensing compressor, then you probably want to compute the combined loudness by taking the vector norm (root of sum of squares, sqrt(L*L+R*R+C*C+...)) instead, because this estimates the combined power (as an approximation for loudness).

Combining a peak sensing compressor with a vector norm over channels, or an RMS sensing detector with a maximum channel makes a lot less sense, because you're basically treating the envelope and the channel link in different ways. There's no rule against doing it... but like you probably get better results if the detector and the channel linking scheme are doing the same thing.

Well, in a sense you are right, from a mathematical perspective and given how a panner works.
I was shortly thinking about this. Let's switch to the stereo case.

But now let's consider the case when you have one guitar left and no guitar right, and then the second rhythm guitar comes in. In a metal song I want the stereo case to bang as hard as possible, so no reduction makes sense from a rather musical perspective.

A peak compressor with max metric will not reduce the level when the second guitar comes in. But the rms compressor would reduce it according to your summing (which by the way makes most sense when both channels are uncorrelated which is rarely the case since bass and kick are center panned).

Now an rms compressor with max metric would behave like the peak one except that it measures the loudness and not peak of each side. So while at first it seams wrong, it has it's point as the loudest signal in any channel drives the compressor. For the peak compressor the result is similar but a bit erratic since the crest factor can vary.

Nothing is perfect. For example, when a mono signal is panned center, the level is typically reduced by 3 dB. Therefore, a peak compressor will compress hard panned signals more, they see a 3 dB lower threshold. But that's how the SSL bus comp seems to work.

In addition to that, when a stereo signal is made mono by summing, the hard panned signals are 3 dB lower than the center panned ones (in LCR mixing). Weird stuff.

For a surround case you may be right with the summing though.

[mystran]

But now let's consider the case when you have one guitar left and no guitar right, and then the second rhythm guitar comes in. In a metal song I want the stereo case to bang as hard as possible, so no reduction makes sense from a rather musical perspective.

A peak compressor with max metric will not reduce the level when the second guitar comes in. But the rms compressor would reduce it according to your summing (which by the way makes most sense when both channels are uncorrelated which is rarely the case since bass and kick are center panned).

This goes a bit more into mixing than plugin design as such, but I would question several assumptions here. First, the idea that you want guitars in a metal song "as hard as possible" is probably not right, because trying to push them too hard just means you'll have to compress so aggressively that you'll end up reducing their impact. That's actually why metal is sort of the "loser" in the "loudness wars" 'cos you absolutely can't push the loudness as high as you could with say EDM without it becoming just a sausage. I listen to a lot of metal actually and the most impactful mixes usually work on maximizing clarity and impact more so than loudness, so that you want to turn the playback volume up... where as with sausage the opposite often happens, where it becomes fatiqueing to listen to and you tend to just either turn the volume down or switch to something else.

The other thing though is do we actually want to handle this sort of hard-panned situation with stereo link or even bus-compression for that matter. The point of stereo link is that since the position of a sound in a stereo field is encoded into the panning, if you then go reducing one channel but not the other, the relative gains of any given sound on the two stereo channels change, hence the sound source moves in the stereo field... but this is a non-issue with hard panned sound because the gain on the other channel is zero anyway.

I'd argue that the case of guitars here is an example where the right thing to do is probably to apply compression (perhaps in mono) individually on each of the guitar tracks and then hard pan them into the mix... or if you want to pan them together first, then apply the compression without stereo link. They are completely separate tracks, on their own individual channels, why should one of them modulate the gain of the other.

And... this is why I think it usually makes sense to pair peak+max and RMS+norm: it's to do with what the stereo link is supposed to do in each of these cases. If we are doing peak control then we essentially MUST use maximum if we want to stereo link (and this is not always given) in order to actually control those peaks effectively... but if we're trying to control loudness with some sort of RMS based approach, then maximum makes little sense, because it means you're basically penalizing (treating as louder) sounds that are panned harder towards a single channel.

Either way, usually you would want to do more drastic compression on the individual tracks so that when they are mixed together, so that the bus compression can act more gently like a glue and perhaps tame the worst variation in dynamics, 'cos this tends to result in much more stable mixes overall, where the drums don't suddenly become weaker (noticeably) weaker just because the guitars came in (or whatever; point is.. you don't usually want a metal mix to sound like some EDM side-chain compression thing where the whole mix breathes).

[mystran]

With the risk of going a tiny bit off topic, I think some of the more "progressive" (and also "symphonic") metal bands have really the "loudness" thing nailed down pretty well.. like you set the overall ambient with the super wide, heavily distorted rythm guitars... but those are not necessarily the loudest part, they are more like just the background.. and then you drop a bright, relatively clean synth lead or perhaps a singer with a powerful voice (often a tenor/soprano, whether it's Eric Adams or Björn Strid or Floor Jansen or whoever) singing clean(!) to really pierce through the wall of sound and maximize the impact... and it's the lead that loud and it can be loud 'cos it's clean, but it makes the whole mix sound powerful and makes you want to turn the gain up at playback precisely because of the impact between the contrasting elements more so than the actual LUFS value printed onto the track.

To bring this back on topic.. I guess the point I'm trying to make is that thinking about compression as something you slap on bus to maximize the loudness is just wrong on every level and it really makes a lot more sense to design a compressor with the idea of "how is this going to help one shape the dynamics into something that works" (rather than something loud). Are you taming track dynamics? Are you enhancing transients? Adding glue? Gently reducing the overall dynamic range to make the track more even so you can play it louder without it suddenly brusting everyone's eardrums?

Sausages are easy. Clear, high impact tracks that you want to play loud are much more difficult.

[Tone2 Synthesizers]

Hi

For stereo one would usually take the maximum of left and right channel OR sum them (mid signal) to get a common detector signal, or doing mid/side compression.

But what about surround when doing a master compression?

I am not sure whether the same approach works here, probably not. Left/Right would likely be treated like in stereo or also including the center channel, what about the the bass channel? Also, perhaps the speakers left and right behind are compressed separately ...

Any info? Thanks

What kind of audio material do you want to compress? Music or cinematic stuff like effects?

What Surround mode do you want to use? Dolby Prologic II (phase encoded to a stereo track) or Dolby Digital / DTS ?