Every other explanation of bus compression hands you a drawing of it. This one makes
sound. Pick a song, hear what a professional engineer actually dialed in on it, then
recreate it yourself on the same record with the same box.
A professional engineer started here and dialed it in to here. Your turn.
Step one
Pick a song
My mix with exactly three things taken off: the bus compression, the parallel Mix B and the
final limiter. Everything else is still on it. This isn't a stripped mix, it's the one
layer you're about to put back.
One song is ready. The other four are being sourced,
one per genre.
Step two
Three settings, one song
Same engineer, same box, same record, three different jobs.
Gentle
Preset one
A couple of dB. You'd miss it if it were gone, and you'd not call it compression if you heard it alone.
Don't start gentle. Get 5 to 10 dB first and make it sound good there, then raise the threshold until it's barely working and still sounds good. You land somewhere different than if you had started soft. His own note: "I'd rather not use bus compression in that case, and I'd rather use heavier parallel compression."
Family
VCA
Threshold
-18.0 dB
Ratio
2.0:1
Attack
30 ms
Release
400 ms
Sidechain
60 Hz
Makeup
+2.5 dB
Moves the needle about 2 dB, peaking near 4.
The SSL grip
Preset two
The one people mean when they say bus compression. It grabs the record and holds it together. Heavy, in his definition, is 4 dB or more.
Two ways to soften it, and they aren't the same. Pull the mix knob back, or dial the reduction back. Try both and keep what sounds better.
Family
VCA
Threshold
-19.0 dB
Ratio
4.0:1
Attack
15 ms
Release
450 ms
Sidechain
80 Hz
Makeup
+4.5 dB
Moves the needle about 4 dB, peaking near 7.
Slammed for parallel
Preset three
Hit far harder than taste allows, then blend it under the mix instead of replacing it. This is the one he actually prefers over gentle bus compression.
It plays on the desk below, not here. A slam belongs on a parallel copy, never across your whole two bus. Pressing it takes you down there to hear the smashed copy alone first.
Family
FET
Threshold
-30.0 dB
Ratio
10.0:1
Attack
2 ms
Release
80 ms
Sidechain
Off
Blend
By ear
Moves the needle about 10 dB, peaking near 14.
GlueBus compressor · The Mix Academy
Nothing loaded yet. Press Hear it on one of
the three cards above, or just start dragging.
Your turn. Six things to match: the family, the
threshold, the ratio, the attack, the release and the sidechain.
His numbers are hidden while you work. Play the
song, watch the needle, and let the chips under each knob steer you.
It listensto the whole mix, low end and all
It decideshow much of the song is too loud
It turns downnothing yet. Press play
That's the whole machine. Everything you loosen or tighten
below changes one of those three things, and the needle underneath shows the third one
happening in real time.
What came off this mix, and what's still on it.
Off: the bus compression, the parallel Mix B and the final limiter, with the gain
backed off to leave headroom. Still on: everything else the mix normally carries, so
it should already sound like a record rather than a pile of parts. That's the point.
You're adding the missing layer, not rescuing something broken.
The evidence that nothing limited it: padded peak -4.78 dBFS, crest factor 15.3 dB,
and no samples at full scale. The compressor here teaches the behavior. Your own
bus compressor supplies the color, because this one is an archetype, not a copy of
any real unit.
Feed-forward and predictable: the attack and release
knobs do exactly what they say. The bus-glue archetype, like an SSL-style bus compressor.
Reading the needle
It sits at zero until the song gets loud enough to
cross the threshold. Then it falls, and how far it falls is how much the compressor
is holding the mix down at that instant.
Barely movinggluing, and you'd miss it if it stopped
Around 4 to 6the grip most people mean by bus compression
Past 10an effect now, not a polish. Fine on a parallel copy
Before the slider
Two paths, one compressor
the audio you hear, and the copy it listens to. Not the same thing.
Trace a path
One signal, two jobs. Only the right-hand copy meets the filter.
Input
splits in two
Audio path
Detector path
Sidechain filteronly ever on this copy
Level detector
audio
a decision, never audio
Gain stagewhere the two meet
Output, what you hear
The filter changes what it reacts to, never what comes out. Reduce the low end in the sidechain and the compressor stops reacting to all of it, so you keep the fullness of the bottom and still get a balanced compression.
Low end carries the most energy, so an unfiltered detector spends nearly all of its
attention down there. On a kick-forward mix the kick makes almost every decision. Raise
the filter and the reduction spreads across the record instead.
The slider below is real. What dims on the spectrum is exactly what the detector
stops hearing.
Stop the kick from making every decisionreduce the low end in the compressor's sidechain and it stops reacting to all of it
Off. The detector hears everything, low end
included, so on a kick-heavy mix like this one the kick makes most of the decisions
and the needle mostly twitches on the kick alone.
ThresholdHow loud before it reacts
RatioHow hard it pushes back
AttackHow quickly it grabs
ReleaseHow quickly it lets go
MakeupAdd back what you took
MixDry against compressed
Level matching is on, so the A/B is a fair fight for
your ears. The needle always shows the raw reduction, before makeup.
Keep what you dialedPlay it, move things until it sounds right, then save the spot. Everything you
save stacks up here and copies out in one go.
Copies as plain text. Nothing leaves this page on its own.
Step three
Why those settings
Not what to set. How he got there, knob by knob, on the preset you
have loaded. The numbers are scaffolding for the ears, never the point.
Load a preset aboveThen this fills in with the reasoning behind each of its numbers.
His reasoning for each number is held back
while you're working, because half of it quotes the numbers themselves. Press
Show me his settings when you want it, or come back to it after your own go.
The gain element
The bus-glue archetype. Feed-forward and predictable, so the attack and release knobs do exactly what they say. This is the family most engineers picture when they say "bus compressor."
How loud before it reacts
Threshold isn't a tone control. It only decides how much of the song is above the line, and therefore how much reduction you get. It's the knob you use last, to back the whole thing off.
How hard it pushes back
How hard it pushes back once it's over the line.
How quickly it grabs
How long the transient gets to escape before the compressor grabs. Slower attack keeps the front of the snare and the kick intact. Faster attack catches them, which is how you control transients rather than glue.
How quickly it lets go
How fast it lets go. This is the knob that makes a bus compressor breathe with the song instead of sitting on it.
What the detector listens to
Filters what the detector listens to, never what you hear. Without it the kick makes almost every decision on a mix like this one, and the rest of the record gets ducked by something it has nothing to do with.
Why the target moves, which is the real lesson
Controlling transients. Faster attack, and a release timed so the snare and
the like behave. Here the compressor is a leveller, not glue.
Capturing the rhythm of the track. Time the attack and release so the
compression moves and reacts with the kick and the snare, and sometimes the vocal,
depending on the balances. The song pumps in time instead of at random.
A lifeless track that needs character. Nothing in it's compressed and it
needs amping up. Hit it harder, and try different compressor families for different
shapes and tones. The four badges at the top of the instrument are that experiment.
Real gear
The families, with real names and real dates on them
The badges on the instrument are an archetype on purpose, so the lesson stays fair to
whatever you own. Here is where those archetypes actually came from: the gain element each
family uses to turn the sound down, the company that built it, and what you can load in
your DAW to hear it.
First axis, the gain element
Feed-forward and predictable: the attack and release
knobs do exactly what they say. The bus-glue archetype, like an SSL-style bus compressor.
Five units, four companies, and the family most bus compression comes from.
160dbx
1976
Screenshot slot. Not supplied yet.
The first VCA clean enough for professional work, built on David Blackmer's own gain cell. Punchy and a little aggressive, which is why it still lives on drums.
In your DAWUAD dbx 160Waves dbx 160
4000 G bus compressorSolid State Logic
1976
Screenshot slot. Not supplied yet.
Built into the centre of SSL's own console and not sold as a box until 1994. In 1978 Townhouse Studios got tired of waiting and built their own from SSL parts.
In your DAWbx_townhousemy templateSoftube Bus ProcessormineUAD SSL 4000 GWaves SSL G-MasterCytomic The Glue
2500API
2002
Screenshot slot. Not supplied yet.
Discrete VCA with the detector on a switch. Old listens after the gain stage, New listens before it. One box, two feels, which makes it a good teacher.
In your DAWWaves API 2500
VSC-2Vertigo Sound
2007
Screenshot slot. Not supplied yet.
Four hand built discrete VCA blocks, with one riding the sidechain as well, so it stays clean even on out of phase peaks. One of the cleanest ever built.
In your DAWVertigo VSC-2, Plugin Alliancemine
VSC-3Vertigo Sound
2019
Screenshot slot. Not supplied yet.
Not an upgrade of the VSC-2 but a different quad discrete VCA design, with peak and RMS switching and a parallel blend built in rather than bolted on.
In your DAWVertigo Sound VSC-3mine
Two units. The fast family, and the one I reach for when I want parallel slam rather than glue.
1176UREI, designed by Bill Putnam
1967
Screenshot slot. Not supplied yet.
Attack down to twenty microseconds and a knee that turns to butter once you push it. Thirteen documented revisions. The all buttons in trick is this box.
In your DAWUAD 1176I slam with thisWaves CLA-76Arturia Comp FET-76Softube FET CompressorAnalog Obsession FETISH, free
Distressor EL8Empirical Labs, designed by Dave Derr
1996
Screenshot slot. Not supplied yet.
Dave Derr took what he liked from the 1176 and the LA-2A and built his own thing instead of a copy. Over forty thousand sold, and a TEC Hall of Fame entry.
In your DAWEmpirical Labs ArousorUAD EL8 DistressorSlate FG-Stress
Three units. A lamp and a photocell do the timing, which is why two of the knobs stop mattering.
LA-2ATeletronix, designed by Jim Lawrence
1965
Screenshot slot. Not supplied yet.
A panel of light shining on a photocell does the timing, so attack and release are that cell's physics rather than a knob. That is the whole reason it forgives you.
In your DAWUAD LA-2A CollectionWaves CLA-2AAnalog Obsession LALA, free
LA-3AUREI
1969
Screenshot slot. Not supplied yet.
The solid state answer to the LA-2A. Same optical cell with 1176 style circuitry around it, so the attack now changes with how hard you hit it. Faster, and ruder.
In your DAWUAD Teletronix LA-3A
CL 1BTube-Tech, designed by John G. Petersen
1991
Screenshot slot. Not supplied yet.
An all tube optical design and the modern reference for a smooth vocal. Sources disagree on the year. Most say 1991, a few say the late eighties.
In your DAWSoftube Tube-Tech CL 1BUAD Tube-Tech CL 1B MkII
Three units, and the oldest gear on this page. The ratio itself rises as you push, so it thickens more than it controls.
STA-LevelGates Radio Company
1956
Screenshot slot. Not supplied yet.
The oldest box here, built to ride levels on AM radio rather than to make records. Very soft knee, very slow release, and people still build clones of it.
In your DAWMixWave Level Compressor
660 and 670Fairchild, designed by Rein Narma
1959
Screenshot slot. Not supplied yet.
Six stages of tube gain reduction. The first unit went to Rudy Van Gelder and the third to Les Paul. Still the thing mastering engineers measure against.
In your DAWUAD Fairchild 660/670Waves PuigChild
Variable MuManley Laboratories
1994
Screenshot slot. Not supplied yet.
Two triodes rebiased by the sidechain. Prized in mastering because it is still doing something musical at one decibel, where most compressors are doing nothing.
In your DAWUAD Manley Variable MuNI Vari Comp
Two units, both Neve, and the family that gets left off most lists including the one this page shipped with.
2254Neve, with David Rees
1969
Screenshot slot. Not supplied yet.
Rupert Neve used the diodes' own changing resistance as the gain element. Wiring them as a bridge is what killed the distortion a single diode would have made.
In your DAWUAD Neve 2254/E
33609Neve
1985
Screenshot slot. Not supplied yet.
The rack version, and the diode bridge everyone actually knows. No VCA, no FET and no light cell anywhere in it, which is why it does not sound like the others.
In your DAWUAD Neve 33609/CArturia Comp DIODE-609IK Multimedia Precision Comp
Second axis, the detector
Where the compressor listens
This is not a sixth family. Any of the five above can be built either way, and where the
detector taps is a bigger part of how a box feels than most people expect.
Feedforward
The detector taps before the gain stage, so it sees the full uncompressed signal and
reacts to that. Most modern designs work this way, the SSL bus compressor included.
Feedback
The detector taps after the gain stage, so it is listening to what it already did and
correcting itself. Older designs, and part of why the 1176 and the 2254 feel as they do.
The API 2500 puts both on one switch, so you can hear the whole
difference on one piece of hardware without changing anything else.
Third axis, the job
Things that are not families at all
You will meet every one of these by name, and not one of them is a gain element. Each can
be built out of any of the five families above, which is exactly why they sit on their own
row instead of in the tabs.
MultibandThe signal is split into frequency bands, each band gets its own compressor, then they are summed back.
Dynamic EQThe input triggers specific EQ bands to move, without splitting the whole signal into filtered bands first.
LimitingA very high ratio with the threshold up near the ceiling. Same mechanism, taken to its extreme.
ClippingNot a compressor. It flattens the tops of the waveform, which is a different operation entirely.
Transient shapingNot threshold based. It responds to how fast the attack is, and moves attack and sustain separately.
ParallelBlending a compressed copy back with the untouched one. A technique, not a circuit.
Peak or RMS, hard or soft kneeWhat the detector responds to, and how abruptly the compression arrives once you cross the line.
One more gain element you will see the name of: PWM, which
switches the audio on and off very fast and sets the level by how long it stays on. Rare
enough that you may never meet one, common enough to be worth the name.
Two names that do not sit in any one row: the Chandler TG1 and
Curve Bender, whose lineage runs back to the EMI console at Abbey Road, and the one off
Townhouse box above, which was never a product at all.
Parallel SlamParallel bus compression · The Mix Academy
Mix A is the mix, untouched. Mix B is the same mix run through
its own compressor, squashed on purpose, with its own fader. Both play together, live,
so you can hear exactly how much of the smash is actually in the blend.
This browser couldn't load the compressor
engine, so Play won't do anything here.
Mix Athe mix itself, untouched
Start with Mix B on its own. Get the smashed copy
sounding great by itself, then blend it under the mix. That order is the whole trick,
because a parallel copy that sounds wrong alone still sounds wrong underneath.
ThresholdMix B only
RatioMix B only
AttackFET clamps this hard
ReleaseMix B only
How hard the copy is being squashed
05162024 dB
Nothing yet. Press play and pull the
threshold down until the marker moves into the blue.
The white line is where my own setting lands, about 13 dB. Measured on this
song through this compressor, off this meter, not a rule of thumb.
Where this came from
Nobody invented it. It got named.
1965
Dolby A noise reduction already ran compression on one of two
parallel paths and summed them back. The shape existed inside a product before
anyone used it as a mixing move.
1977
Mike Beville writes it down in Studio Sound, applied to
classical recording. He calls it side-chain compression, which confused people for
years, because that phrase already means something else.
1980s
Bob Clearmountain leans on it across a decade of rock records
and it stops being a curiosity.
2002
Bob Katz gives it the name parallel compression in
Mastering Audio, and frames it as upward compression: lifting the quiet rather than
taming the loud.
the name
New York compression comes from the reputation of that
city's studios, not from one engineer. Nobody owns it, which is probably why it
spread.
today
Kenny Gioia is who most people actually learn it from.
On a real mix this is rarely one compressor. I will often stack two or three, sometimes a limiter after them, each one doing a little of the work so the squashed copy still sounds good squashed.
Makeup gain lives inside the boxIt puts back the level the compression just took, so you can judge the sound
instead of the volume drop. It changes nothing about how hard the copy is squashed.
Output gain is the fader on the rightIt decides how much of the squashed copy joins the mix. This is the one that
makes parallel compression parallel, and it is the last thing you should touch.
That is the art of it: not just hitting it hard, but making it sound right while it is hit hard. One box here keeps the lesson clean. The fader on the right still decides how much of it you keep.
Mix Bthe smashed parallel copy
There's no right number for the Mix B fader. It's song
dependent. Mute it, listen, unmute it, and keep whatever made the mix sound better.
1of 9
Pick a song
Everything below runs on whatever you pick here.
Thresh-20.0 dBRatio4.0:1Attack30 msRelease300 msSidechainOffMix100%Reduction0.0 dB