KVR Audio

Franzel · Post by **Franzel** » Wed Apr 01, 2009 6:51 pm

andy_cytomic wrote:Thanks to everyone that has been helping out with their comments about The Glue and testing it out and generally putting up with a few bugs. With your help I've been able to improve the plugin and I hope now to really be able to take on the big boys

Andrew Simper

Thanks for the great job you did on The Glue Andy.

Keep up the great work, guess you already gave the big boys something to chew on

Cheers, Hans

djsubject · Post by **djsubject** » Wed Apr 01, 2009 7:01 pm

just testing now on some drums

i like the slow attack much better now!!

or its a placebo April first joke, in witch case, you got me

Subz

Morris · Post by **Morris** » Wed Apr 01, 2009 7:30 pm

Thanks Andy, for making a hell of a Glue. At the moment one of the best compressors for me!!!! Have a nice time @ the messe.

mystran · Post by **mystran** » Wed Apr 01, 2009 8:36 pm

First thought I'd not poison this thread, but I had adl run me the stereotest that I posted through UAD.. and then informed him about the result, and he suggested that I'd tell andy, and I mailed andy (through the kvr email link) too... but then I couldn't resist using the new 1.0.10 renewed demo period to run the same test through the Glue as well.. and then after thinking about the results I realized there would be way to convert one behavior into the other.. so I thought I'd be useful to post it anyway:

So the difference in how these two plugins handle stereo summing (in static case anyway, and accounting for 10:1 ratio not being total brickwalling):

UAD either runs on L/R average (which corresponds to maximum of mid and side) or has a mono side chain (I only afterwards realized I forgot to include test signal for that, but if stuff isn't overpanned beyond hard-left/hard-right it doesn't matter). Equal power (-3dB center panning law) left-panned signal comes out of the compressor with about 6dB higher peak than mid-panned signal (where as going into the compressor the difference is 3dB).

Glue runs on maximum of left/right (which corresponds to averaging mid and side). Equal power left-panned comes out with essentially the same peaks as mid-panned (so you lose the 3dB difference in peaks).

Practical consequences: UAD keeps stereo information consistent with respect to -6dB center panning law, where as Glue keeps it consistent with respect to 0dB center panning law. With respect to -3dB center (equal power or circular) panning law, UAD will compress left/right panned signals less, and Glue more.. hence UAD will "widen", and Glue "narrow" signals (with respect to circular panning) when compressing mixed stereo contents; neither of them "tilts" the stereo image though.. they apply the same amount of compression to both channels. Just different detection behavior in respect to stereo.

Conversion: if UAD has stereo side-chain (that is, can act on pure stereo material, really really sorry for forgetting that from the test) you can convert it's (static) behavior into Glue's (static) behavior by sticking an M/S encoder before the plugin, and M/S decoder afterwards (if the side-chain is mono, that won't work, again, sorry for realizing the lacking test signal only afterwards)...

For Glue, you can get the UAD results by doing the same MS encoding/decoding, or use the external side-chain and feed it M/S encoded version of the signal (or if UAD happens to have mono side-chain, that result can be had by feeding Glue's side-chain with mono-sum of the input). Note that you might need to tweak threshold afterwards in both cases to get the same compression.

There can obviously be any number of other differences with respect to side-chain operation, which are not addressed by this test, which only tests level detection from stereo in the static case, because that really was the only thing I was interested.

Bottom line: it's not the same, but you can have the results anyway with the external side-chain input of the Glue (or just MS encoding/decoding before/after the plugin).

penguinfromdeep · Post by **penguinfromdeep** » Wed Apr 01, 2009 10:45 pm

You're just splitting some hairs there mystran

Seriously it was fun read, thanks

scoobz · Post by **scoobz** » Wed Apr 01, 2009 10:58 pm

I thought it was an interesting bit of detective work!

mystran · Post by **mystran** » Wed Apr 01, 2009 11:02 pm

penguinfromdeep wrote:You're just splitting some hairs there mystran Seriously it was fun read, thanks

For the purpose of slight compression on a mostly mono track, yeah I'm splitting hairs. For the purpose of aggressive smashing of stuff panned wildly all around the stereo field, we're talking up to 6dB differences (that's twice the amplitude, or 4 times the power) if the compressor is set fast enough to fully react.

gutterman · Post by **gutterman** » Thu Apr 02, 2009 1:38 am

AUTO-ADMIN: Non-MP3, WAV, OGG, SoundCloud, YouTube, Vimeo, Twitter and Facebook links in this post have been protected automatically. Once the member reaches 5 posts the links will function as normal.

wow!just wow! ive spent the half of the night working on a dry example of andy's nefreonsurfer120comp( http://www.cytomic.com/AudioExamples/Th ... lectronic/ (http://www.cytomic.com/AudioExamples/TheGlue/Electronic/) ) with my waves sslgbuss and couldnot recreate this phatness!what settings did you use on that one?if i wont recreate it with my waves ill buy the Glue at the very same moment!

drillbit · Post by **drillbit** » Thu Apr 02, 2009 2:52 am

Mystran, I am very interested to hear Andy's comments on your stereo field research. I know he is very particular about stereo field and ensuring it is not effected by the compressor. Does the UAD plugin increase stereo field (turn down the center channel in relation)? That would a be a bad thing. I hear no decrease in stereo field with The Glue in headphones or on my Adams.

Interesting stuff. Now to really try out this new algorithm. I'm a bit concerned about it actually as I have used The Glue so extensively.

mystran · Post by **mystran** » Thu Apr 02, 2009 3:39 am

drillbit wrote:Mystran, I am very interested to hear Andy's comments on your stereo field research. I know he is very particular about stereo field and ensuring it is not effected by the compressor. Does the UAD plugin increase stereo field (turn down the center channel in relation)? That would a be a bad thing. I hear no decrease in stereo field with The Glue in headphones or on my Adams.

Just to clarify: first of all I don't have the UAD. I had someone run a simple test for me. I'd have to ask for more tests to come up with a lot of detailed information about it's behavior, and frankly I couldn't care less. Second, neither plugin (as far as I can tell) does anything with respect to mid/side balance (stereo field width in the traditional sense) when the compression gain remains constant. Actually screwing this up is pretty hard, unless you actually want to do dual-mono or separated mid/side processing.

The difference is that if you have a sound with certain level (let's they say are same total power for -3dB center panning, but the difference remains the same even if not) panned middle, followed by another sound panned to one of the sides (not side-channel, but simply hard left or right) then UAD seems to detect the hard-panned sound as if it was 3dB lower than the mid panned sound, where as Glue seems to detect the the hard-panned sound as if it was 3dB higher than the mid panned sound (and when some other compressor plugins let you choose between stereo average or stereo max, UAD would be the average case, and Glue would be the max case).

Put another way, the parts of the track where the stereo correlation is very high (the signal is mostly mono) will get compressed more by UAD than the parts where stereo correlation is lower (at least before it goes into anti-correlation which would be the overpanned case that I can't comment on, because I overlooked it). This results in the low correlation (wide) parts remaining louder (less compressed) which can give an impression of widening... where as the Glue by default seems to do the opposite: parts where the stereo correlation is lower, get compressed more.

Don't know what the original hardware does (I could search back for the schematic that andy linked, but I guess it does what the Glue does). Just wanted to point that the people that thought UAD was wider weren't wrong: it really pulls low-stereo correlation (wide) parts of tracks higher in relation to high-correlation parts (say kick hits or whatever).

Oh and the behavior that andy has is optimal with regards to getting a signal hot without hitting the clipper, because hard-panned left/right sounds will hit clipper first.. in case you were trying to use the Glue as a limiter.

If you assume humans sense signal power (which might or might not be a good assumption, and ignoring any complications with acoustic summing) then neither really does the right thing which I guess is what dbx guys call True Power Summing (notice that it's their capitalization not mine.. not sure if there should be a trademark superscript too)...

As for saying which one is best? Well it's exactly the same question of which panning law you think is "correct": UAD gives you results consistent with -6dB center, Glue gives you results consistent with 0dB center... and if you picked pythagorean mean sqrt(l*l+r*r) to detect against, you'd get -3dB center. A matter of taste I guess..

Oh and maybe they aren't purely sum vs. average and one or the other them is doing something really fancy. But they seem to be close enough that I don't personally have a reason to assume otherwise.

scoobz · Post by **scoobz** » Thu Apr 02, 2009 4:05 am

Another interesting read thanks. So essentially you're saying that even if two compressors are identical, if the way they sense the information is different they will give a different output manifesting in a perceived widening or focusing (depending on your perspective) effect?

So if the glue were to have a controllable sensing option (in relation to m/s info) it would be approaching 'the perfect compressor'?

(whatever that is!

)

mystran · Post by **mystran** » Thu Apr 02, 2009 4:51 am

scoobz wrote:Another interesting read thanks. So essentially you're saying that even if two compressors are identical, if the way they sense the information is different they will give a different output manifesting in a perceived widening or focusing (depending on your perspective) effect?

Yeah, kinda.. because the compressor is flattening the signal (towards a situation where the signal always has the same gain) it kinda needs to consider how "same" gain is defined for stereo information (which for pan-potted mono is known as the panning law) when it senses the original dynamics.

There's a number of other possible differences that my test didn't even attempt to measure such as whether you detect first and sum afterwards or the other way around, but being able to say anything meaningful about such things would then require analysing the whole detection method, which I personally have no interest into.

So if the glue were to have a controllable sensing option (in relation to m/s info) it would be approaching 'the perfect compressor'? (whatever that is! )

I don't know about perfect. But route the Glue's side-chain through M/S encoder and you should kinda have the -6dB panning law (which can result in a bit wider result) if you happen to prefer that.

Maybe for perfect compressor we need the other possibilities too? If you norm the L/R vector with (abs(l)^x + abs(r)^x)^(1/x) you get -6dB for x=1 and 0dB for x=infinity. For -3dB you'd use x=2. For other values between 0dB and -6dB use other values between 1 and infinity.. for center attenuation more than -6dB I guess you could use x<1 (though I think mathematicians insist that's not a norm). For bumping the center (not sure why you'd ever want to do that) you can't easily go above infinity, but you can swap the basis to M/S field and use (abs(m)^x + abs(l)^x)^(1/x), where infinity norm (max) becomes straight sum (x=1) and straight sum becomes the x=infinity case (everything kinda turns around) at which point you should be able to use x<0 values for a center amplifying panning law.

Note though, that the above assumes mid-channel is handled the same as side-channel (so if the mid-channel is -6dB with respect to left or right, the pure side-channel when left/right are anti-phase to each other is also -6dB with respect to left or right channel).

But it's only a difference in level detection. With constant detected gain stuff will stay panned the same as it was originally, and no widening or narrowing will ever happen as long as you apply the same gain to both channels (whether L/R or M/S).

Audio~Geek · Post by **Audio~Geek** » Thu Apr 02, 2009 4:54 am

This is a damn fine compressor.

Where's that link again? (not on page 1, not in Cytomic's sig) ¯\(°_o)/¯

Don't make me google it.

mystran · Post by **mystran** » Thu Apr 02, 2009 4:57 am

Audio~Geek wrote:This is a damn fine compressor.

Where's that link again? (not on page 1, not in Cytomic's sig) ¯\(°_o)/¯

Don't make me google it.

Googled for you: http://www.cytomic.com/Beta/The_Glue_Beta_Info.html

adl · Post by **adl** » Thu Apr 02, 2009 5:44 am

@mystran
Hi there! Just wanted to ask, if you could try again with the new 1.0.10 as i think that "the glue" has really improved with this new release. Andy did some reprogramming / algorithm changes. I think it also effects the stereo behavior in a good way (couldn't tell any difference between "the glue" and the UAD 4K anymore).
Could you run the test again with the new version, please?

Thanx!
Thomas

Cytomic 'The Glue' Compressor