Audio Compressor

Audio Compressor sits anywhere in an audio chain: source → filters → Audio Output, and is a natural last-in-chain node before the output. Connecting the Audio input splices it into the shared AVAudioEngine graph — the cables are the engine wiring, one clock drives everything, and it can be added to a running chain without a glitch.

Internally the node hosts Apple's DynamicsProcessor Audio Unit, configured as a compressor only. The stock DynamicsProcessor is secretly a compressor plus an expander, and the expander's default settings pull quiet audio down toward silence — the classic "the compressor ate my sound" trap. This node neutralizes the expander outright (ratio 1:1, threshold at −120 dB), so quiet passages pass untouched and only the compressor section acts.

Threshold sets the level (dB) above which gain reduction begins. Headroom shapes the soft knee — the transition zone above the threshold over which compression progressively takes hold: small values give a harder, more audible knee; large values compress so gradually the effect becomes invisible. Attack is how fast the compressor clamps down once the signal crosses the threshold (short = catches transients, long = lets the initial hit through). Release is how fast gain returns to unity afterward (short = punchy but can pump; long = smooth but can duck the material). Makeup adds fixed output gain to restore the loudness that compression removed.

The Audio output re-emits the processed signal as a new engine-backed snapshot, so more filters can follow and an Audio FFT can tap the chain after the compressor — analysis then sees the leveled signal. The Level output is a peak-decay RMS of the compressed signal; comparing it against an Audio Gain's Level before the compressor visualizes the gain reduction. Enable = false engages the audio unit's clean bypass: audio keeps flowing, unprocessed.

Outputs: Audio (the processed signal, chainable) and Level (peak-decay RMS of the compressed signal).

Musical starting points: gentle mix glue = Threshold −15, Headroom 15, Attack 0.01, Release 0.3; drum punch = Threshold −20, Headroom 3, Attack 0.03, Release 0.08; broadcast voice = Threshold −25, Headroom 8, Attack 0.002, Release 0.15, Makeup +6.

• Mix-bus glue before output: Audio Player → Audio Equalizer → Audio Compressor (Threshold −15, Headroom 15, Release 0.3) → Audio Output — a couple of dB of invisible leveling on the whole chain.
• Steady live-mic voice: Audio Input (mic) → Audio Compressor (Threshold −25, Attack 0.002, Makeup +6) → Audio Reverb → Audio Output — consistent speech level regardless of mic distance, on live input via the engine's realtime source bridge.
• Delay/reverb tail safety net: Audio Player → Audio Delay (high Feedback) → Audio Compressor → Audio Output — the compressor catches runaway feedback build-up before it hits the speakers.
• Stable analysis signal: Audio Player → Audio Compressor → Audio Output, with an Audio FFT tapped after the compressor — leveled dynamics keep spectrum visuals consistently energetic through quiet and loud sections alike.
• Punch shaper for drums: Audio Player (drums) → Audio Compressor (Attack 0.03, Release 0.08, Headroom 3) → Audio Output — the slow attack lets each hit's transient snap through, then squeezes the body.
• Level-matched distortion taming: Audio Player → Audio Distortion → Audio Compressor (Threshold −10, Makeup −3) → Audio Output — reins in the loudness jump a hot preset creates.

• Tune to the program material: percussive sources want slower attacks (0.02–0.05 s) so transients survive; voices want fast attacks (1–3 ms); full mixes want wide Headroom (10–20 dB) and slower releases so the compression disappears.
• If the sound audibly "pumps", lengthen Release; if it feels lifeless and flattened, raise Threshold, widen Headroom, or slow Attack — over-compression is the most common mistake.
• Always level-match when judging: add Makeup until bypassed (Enable = false) and engaged sound equally loud — louder always sounds "better", which fools the ear.
• The hidden expander that makes Apple's DynamicsProcessor swallow quiet audio is neutralized in this node — quiet passages pass at unity, so what you set is what you get.
• Compressor placement: after distortion tames its loudness spike; after delay/reverb rides the tails; before a reverb keeps the space's input steady. Last-in-chain before Audio Output is the safe default.
• Watch gain reduction visually: tap Level before (via an Audio Gain node) and after the compressor and plot both — when the two traces diverge, that gap is your compression in action.