Compression,
Actually Explained

The transfer curve

The best way to visualize a compressor: input level on the x-axis, output level on the y-axis, both in dBFS. No compression is a straight 1:1 diagonal. Add a threshold and ratio and the line bends: below the threshold, signal passes untouched; above it, the slope flattens. Play with the controls below.

The six parameters

1. Threshold

Signal below the threshold passes untouched. Signal above it gets reduced. Too high and the compressor barely fires. Too low and you're crushing dynamics constantly.

2. Ratio

At 4:1, every 4 dB over the threshold yields only 1 dB out. 2:1 is gentle. 8:1 is limiting territory. ∞:1 is a true hard limiter.

3. Attack

Attack decides whether the compressor preserves or kills the transient. Fast attack (0.1–5 ms) clamps immediately, rounding off the initial peak. Slow attack (20–60 ms) lets the transient through before the compressor engages, so the hit snaps and the compressor works the body (which is usually what you want on snares and drum buses).

4. Release

Release controls how quickly the compressor returns to zero gain reduction. Too fast (10–30 ms) and you hear each compression cycle as a separate artifact. Too slow (500 ms+) and it's still clamping when the next transient arrives.

Release timing relative to tempo:

  BPM    | 1/4 note | 1/8 note | 1/16 note
  -------|----------|----------|----------
   80    |  750 ms  |  375 ms  |  188 ms
  100    |  600 ms  |  300 ms  |  150 ms
  120    |  500 ms  |  250 ms  |  125 ms
  128    |  469 ms  |  234 ms  |  117 ms
  140    |  429 ms  |  214 ms  |  107 ms

Formula: ms per beat = 60000 / BPM
         ms per 16th = 60000 / BPM / 4

5. Knee

Hard knee: ratio switches on instantly at the threshold: nothing below, full ratio above. Soft knee: ratio ramps up gradually over ±3 dB around the threshold. Softer is more transparent. Harder is more precise. Try both in the graph above at 4:1 and −18 dBFS.

6. Makeup gain

Compression pulls the level down. Makeup gain puts it back. Without it, compressed signals sound quieter, but quieter always sounds worse in an A/B, regardless of quality. If you're averaging 6 dB of GR, add about 4–6 dB back. You don't need to compensate 1:1; the peaks are already gone so you have headroom you didn't have before.

Reference settings: four common scenarios

Drum Bus Glue (SSL G-Bus style / Glue Compressor):
  Threshold:  -18 dBFS   catching peaks but preserving body
  Ratio:       2:1        gentle, transparent
  Attack:     30 ms       lets transient snap through
  Release:    auto        program-dependent, ~100-150ms at 120BPM
  Knee:        soft
  Makeup:     +2 dB
  Target GR:  2-4 dB

Vocal Compression Chain (two stages):
  Stage 1 — fast leveler (1176-style FET):
    Threshold:  -14 dBFS
    Ratio:       4:1
    Attack:      3 ms      catches harsh consonants quickly
    Release:    40 ms
    Makeup:     +3 dB
    Target GR:  4-8 dB
  Stage 2 — smooth LA-2A-style optical:
    Works on what the FET missed; optical program-dep. release

Kick Drum (parallel blend):
  Threshold:  -20 dBFS
  Ratio:       6:1
  Attack:     10 ms
  Release:    60 ms
  Knee:        hard
  Blend:      50% wet / 50% dry (preserves original transient)

Master Bus Compression (gentle, transparent):
  Threshold:  -3 to -5 dBFS
  Ratio:       1.5:1
  Attack:     10-30 ms
  Release:    auto
  Knee:        soft
  Target GR:  1-2 dB only — this is glue, not leveling

Detection modes: RMS vs Peak

Peak detection

Responds to instantaneous sample peaks. Fast and precise: it fires the moment the signal crosses the threshold. Right for limiting, de-clicking, and surgical peak control. The 1176 is peak-detecting, which is why you can push it so fast without it sounding erratic.

RMS detection

Averages signal power over a short window (~10–50 ms). Short transients don't trigger it because it responds to energy, not spikes. More musical on complex material, riding average loudness instead of reacting to every hit. The SSL G-Bus and LA-2A both work this way.

Sidechain compression

Every compressor has two signal paths: the audio path (what you hear) and the detection path (what triggers gain reduction). By default they're the same signal. Sidechain routes a different signal into the detection path, so the compressor ducks the audio based on what's happening elsewhere.

The classic kick-sidechain-bass duck

Compressor on the bass channel, sidechain fed by the kick. Every time the kick fires, the bass gets clamped 4–8 dB, clearing space. When the kick is silent, the bass comes back up. That rhythmic breathing is house, techno, trance.

Sidechain Bass Duck — classic house/techno:
  Plugin:     compressor on Bass channel
  Sidechain:  kick drum (often a copy of the kick, pre-effects)
  Threshold:  -20 dBFS  (fires immediately when kick hits)
  Ratio:       4:1 – 8:1 (strong duck)
  Attack:      1-5 ms    (fast — you want the duck to happen with the kick)
  Release:    100-200 ms (controls how long the duck lasts — key parameter)
  Makeup:      0 dB (the duck IS the effect)
  GR target:   6-10 dB  (you should see it clearly on the meter)

  To control pump intensity:
    Longer release = slower breath, more dramatic swell
    Shorter release = snappier, tighter, less pumpy
    Lower threshold = compressor fires harder, bigger duck
    Higher threshold = lighter touch, subtler movement

Controlling the pump

The pump is all release time. At 120 BPM with kicks every 500 ms, a 400 ms release gives you one full breathing cycle per beat. For a locked pump that resolves cleanly before the next kick, set release just under the inter-kick gap. At 120 BPM, 400–450 ms is the sweet spot.

Sidechain frequency shaping:

  Problem: the hi-hats also trigger the sidechain
  because they share the same drum bus feed.

  Solution: high-pass filter the sidechain signal
  so only low-frequency content (the kick's sub
  and body) triggers the compressor.

  In Ableton: use a compressor on the bass track,
  enable external sidechain → kick track. Add an
  EQ on the sidechain monitor path. HPF the sidechain
  at 80-150 Hz to isolate kick body from hat bleed.

Sidechain in the frequency domain

Fabfilter Pro-C 2 and similar compressors let you filter the sidechain by frequency so the compressor only responds to a specific range. Classic use: de-essing (filter to 5–10 kHz so it only fires on sibilance) and blocking low-frequency content from triggering unwanted bus compression.

Parallel compression (New York compression)

Run the signal through a heavily compressed parallel path and blend a portion of that crushed signal back with the dry. The dry signal keeps its transient punch. The compressed signal adds body, density, and sustain. In the gaps between hits (the sustain, the room, the bleed), the compressed path brings up material that was previously inaudible. That's where the density comes from.

Parallel Compression Setup (Ableton / any DAW):

  Option 1 — Send/Return:
    1. Create a Return track
    2. Set the compressor on the Return track:
         Ratio:      8:1 – 20:1 (crush it)
         Threshold:  -24 to -30 dBFS
         Attack:     5-15 ms
         Release:    50-100 ms
         Makeup:     +10 to +15 dB (bring it back up loud)
    3. Send your drum bus at -inf, then raise until
       you hear density without losing snap
    4. Target blend: 20-40% wet (by ear)

  Option 2 — Plugin wet/dry knob:
    Set compressor to 100% compressed internally,
    then use the dry/wet knob to blend.
    Simpler, but less flexible routing.

Parallel vs series compression

Series stacks two compressors sequentially, which works well for two-stage vocal chains where a fast FET catches peaks and a slow optical rides the overall level. Parallel preserves transient character because the dry signal is never touched. Use parallel for density and sustain without losing punch. Use series for two different compression behaviors on the same signal.

Compressor circuit types

Plugin recommendations

Compression in context

Drum bus

Goal: make kick, snare, hats, and percussion move as one object through the mix. Secondary: add rhythmic energy by shaping the whole drum group's envelope.

Drum Bus — Glue Compressor / SSL G-Bus:
  Threshold:  -18 to -22 dBFS
  Ratio:       2:1
  Attack:     30-40 ms     (let the hits breathe)
  Release:    auto or ~120 ms at 120 BPM
  Knee:        soft
  Makeup:     +2 to +3 dB
  GR:          2-5 dB

  Signs you've gone too far:
    - Drum kit sounds flat and lifeless
    - Kick punch disappears
    - Everything sounds equally loud (no dynamics)
    - Mix starts to sound "small"

Stereo / master bus

Goal: cohesion and a consistent loudness feel. Should be nearly inaudible: 1–2 dB of GR that you only notice when you bypass it.

Master Bus — Neve 33609 / SSL G-Bus:
  Threshold:  -3 to -5 dBFS   (firing only on peaks)
  Ratio:       1.5:1
  Attack:     10-30 ms
  Release:    auto
  Knee:        soft
  GR:          1-2 dB maximum

  Working at 1-2dB GR means the compressor is mostly
  not firing — it is a safety net and a subtle glue,
  not an active gain-riding tool.

Bass

Bass Compression (optical or VCA):
  Threshold:  -16 to -22 dBFS
  Ratio:       3:1 – 4:1
  Attack:     10-20 ms     (catch the pluck or synth attack)
  Release:    60-120 ms    (hold through the sustain)
  Knee:        soft
  Makeup:     to taste

  For sub bass in electronic music:
    After compression, add a limiter at -3 dBFS
    to hard-clip any remaining rogue peaks.
    Sub frequencies have huge amplitude swings
    that even a compressor can miss.

Synth leads and pads

On leads, slow attack lets the oscillator's initial edge punch through; fast attack rounds it off. On pads, gentle RMS compression (2:1, slow attack, auto release) acts as a level rider. Keep GR under 3 dB. Pads should breathe.

Vocals in electronic music

Two-stage vocal chain:

  Stage 1 — Fast peak control (1176-style FET):
    Threshold:  -12 to -16 dBFS
    Ratio:       4:1
    Attack:      2-5 ms
    Release:    40-60 ms
    GR:          3-6 dB
    Goal: remove harsh spikes, not level the performance

  Stage 2 — Slow leveler (LA-2A-style optical):
    Threshold:  -20 to -24 dBFS
    Ratio:       3:1 (or optical — no ratio knob)
    Attack:     slow (program-dependent)
    Release:    slow (program-dependent)
    GR:          4-8 dB
    Goal: ride the performance level so every phrase
          sits consistently in the mix

  Optional Stage 3 — Parallel blend:
    Add a crushed version (8:1, -24dB threshold)
    blended at 20-30% to fill in breath and body
    between words.

Reading the gain reduction meter

The GR meter is more useful than the transfer curve visual. It shows in real time exactly how many dB the compressor is pulling down.

Cheat sheet

Ratio: 2:1
Threshold: -18 to -22 dBFS
Attack: 30 ms
Release: auto
Knee: soft
GR: 2-4 dB

SC input: kick track
Ratio: 4:1 – 8:1
Threshold: -20 dBFS
Attack: 1-5 ms
Release: 100-200 ms
GR: 6-10 dB

Stage 1 FET: 4:1, -14dBFS, 3ms atk
Stage 2 Optical: program-dep.
Parallel blend optional at 20-30%

Ratio: 1.5:1
Threshold: -3 to -5 dBFS
Attack: 10-30 ms
Release: auto
GR: 1-2 dB only

Ratio: 8:1 – 20:1
Threshold: -24 to -30 dBFS
Attack: 5-15 ms
Release: 50-100 ms
Blend: 20-40% wet

ms per beat = 60000 / BPM
120 BPM → 500ms/beat
           → 125ms/16th
Set release ≈ 0.8× inter-kick gap

VCA:        fast, transparent, precise
FET:        fast, aggressive, color
Optical:    slow, musical, forgiving
Variable-μ: slowest, warmest, expensive

Mastering:   1-2 dB
Bus glue:    2-4 dB
Vocals:      4-8 dB
Parallel:    10+ dB (blended back)
Limiting:    as needed