PUSH docs / KERN audio

what PUSH does

PUSH is a three-character compressor that works across 40 perceptually-spaced ERB frequency bands. it sits before your final compressor in the chain, shaping a signal so your Pro-C, 1176, LA-2A, or SSL bus comp has less work to do.

three characters, one job. SETTLE for bus glue, TIGHT for transient grip, DRIFT for the thing no other compressor does: per-band upward compression that brings up ambience, room tone, sustain tails, and high-frequency air without pumping up low-frequency mud.

five controls maximum. four knobs (AMOUNT, SPEED, FOCUS, MIX) plus a character selector. the curves underneath are pre-tuned through blind A/B against reference compressors on real source material. you pick a character. you push one knob. the rest stays out of the way.

use it on: drum buses, vocals, mix bus, anywhere a single compressor sounds too one-dimensional. designed to complement your final compressor, not replace it.

controls

PUSH has 4 knobs, a 3-way character selector, a 2-way stereo mode toggle, and three independent buttons for DELTA, AUTO, and SIDECHAIN.

knobs

control	range	default	what it does
`AMOUNT`	0–100%	30%	overall compression amount. scales the gain reduction depth across all 40 bands. on DRIFT, also scales the upward bias (0% AMOUNT = zero gain change).
`SPEED`	0–100%	50%	attack/release timing scale. 0% is slowest (think LA-2A opto), 100% is fastest (think 1176 fast mode). character also influences the curve shape underneath.
`FOCUS`	20 Hz – 20 kHz (log)	1 kHz	frequency-dependent threshold centre. shifts which bands the compressor leans on. low FOCUS = bass-heavy compression, high FOCUS = top-end focused. on DRIFT, also tilts the upward bias toward higher frequencies as FOCUS rises.
`MIX`	0–100%	100%	wet/dry blend. use 30–70% for parallel compression (New York style).

character selector

character	sound
`SETTLE`	soft-knee downward compression with ratio 1.5:1–4:1 and a 20–30 dB knee width. dual-time-constant auto-release (fast per-band + slow broadband, soft-max combined). SSL bus compressor DNA. for glue.
`TIGHT`	medium-knee downward compression with ratio 4:1–12:1 and a 12–3 dB knee width. program-dependent ratio (crest factor modulates it). GR-dependent release that slows during deep compression. 1176 FET DNA. for grip.
`DRIFT`	per-band upward compression with dual-threshold gating. spectral tilt bias toward higher frequencies. two-phase release with explicit 50% recovery crossover. LA-2A T4 opto DNA, applied per-band. for breath, room, and the air no other compressor recovers.

switching characters during playback is click-free. PUSH crossfades between characters over ~10 ms.

toggle buttons

button	what it does
`DELTA`	hear only the gain reduction. output = wet − dry. you hear what PUSH is removing (or, on DRIFT, what it is adding).
`AUTO`	K-weighted loudness compensation (ITU-R BS.1770). on by default. matches output loudness to input loudness so your A/B comparison is honest. you hear the character, not a volume difference.
`SIDECHAIN`	enables external sidechain input. routes the sidechain signal through the same 40-band ERB envelope detector that drives compression. supports stereo and mono sidechain.

stereo mode

mode	what it does
`L/R`	processes left and right channels independently. default. transparent for centred mono sources, slight stereo widening on stereo sources.
`M/S`	encodes the signal to mid/side, processes each separately, decodes back to L/R. useful when you want different compression behaviour on the centre versus the sides. tighter bass without flattening the stereo image.

character deep dive

the three characters are not just timing presets. each uses a fundamentally different gain computer, transient gate, release behaviour, and optional harmonic coloring. switching characters changes how the compressor thinks, not just how it sounds.

SETTLE / bus glue

soft-knee downward compression with a wide ratio range. dual-time-constant auto-release combines a fast per-band envelope follower with a slow broadband envelope, soft-max combined so neither dominates. transient gate reduces gain reduction by 30% during onsets to preserve punch. sidechain HPF rolls off below 200 Hz to prevent kick-dominated pumping. asymmetric tanh coloring adds gentle even harmonics (0.1–0.5% THD), engaged from 3 dB GR upward. this is the closest PUSH gets to SSL G-Bus territory: subtle, musical, holds a bus together without flattening it.

TIGHT / transient grip

medium-knee downward with higher ratios and a much faster envelope. program-dependent ratio means the compressor backs off during transients and tightens during sustain (the crest factor modulates the curve). GR-dependent release slows the recovery during deep compression (models the 1176 feedback topology where heavy GR holds the signal down longer). dynamic odd/even harmonic blend: at low GR, 60/40 odd/even (warm); at deep GR, 85/15 (aggressive FET). transient gate is the most aggressive of the three (85% GR reduction at onset, near-full passthrough on the very fastest material).

DRIFT / breath, air, room

per-band upward compression with dual-threshold gating: signal below threshold is raised, signal above is unaffected. spectral tilt biases the upward gain toward higher frequencies (+4 dB at 20 kHz, -4 dB at 20 Hz, scaled by AMOUNT). per-band history-dependent release: each band accumulates "compression memory" during active use (~600 ms to full warm), decaying back to cold over ~12 seconds. the result is non-stationary: same signal processed repeatedly produces slightly different results. LA-2A T4 thermal-cell DNA, applied per-band.

DRIFT is the structural differentiator. no other compressor at any price point offers per-band upward spectral compression with this gating and tilt model. it is the lever you reach for when a track sounds correct but feels flat: it pulls up the room, the cymbal decay, the breath at the end of a vocal phrase.

key insight: SETTLE and TIGHT are downward (they pull peaks down). DRIFT is upward (it raises detail below threshold). the three characters cover three different problems: glue, grip, and recovery.

spectral display

the real-time spectral display shows:

magenta curve: the post-compression output spectrum, brightly drawn
dim ghost curve: the pre-compression input spectrum, drawn behind the output so you can see what PUSH changed at a glance
peak hold line: a solid trace that trails above the moving spectrum when the signal stops, decaying at 3 dB per second (industry-standard SPAN / Pro-Q style)
transient flash: brief curve brightening on kick, snare, or any percussive hit. baseline-subtracted so sustained music does not trigger false flashes

the pre-curve ghost is the diagnostic. if you see the bright output curve sit lower than the ghost across the spectrum, SETTLE or TIGHT is doing downward compression. if you see the bright curve sit higher (especially toward the top end), DRIFT is doing upward compression. if you see them overlap, MIX or AMOUNT is low and PUSH is barely engaged.

AUTO loudness matching

AUTO is on by default. it applies K-weighted loudness compensation (ITU-R BS.1770) so the output loudness matches the input loudness regardless of how much GR the compressor is doing. when you A/B against bypass, you hear the character of the compression, not a volume difference. turn AUTO off only if you want to hear the raw downward action (the volume drop).

EQ sensitivity nodes

four interactive parametric nodes sit on the spectral display (teal, grey, orange, purple from low to high frequency). they shape how aggressively PUSH compresses each frequency region — think of them as per-band sensitivity, not output EQ. a node with positive gain makes PUSH lean harder on that region. a node with negative gain pulls back.

drag a node to move it. horizontal drag changes frequency, vertical drag changes gain. shift + drag for fine control.
hover a node to reveal inline controls: a clickable enable dot (left of node) and a type badge (for bands 0 and 3 — click the badge to cycle BELL / LS / HS).
double-click the label text ("250 Hz +0.0 dB Q 1.0") to type exact values: "2.5k +3.0 dB Q 0.7". enter commits, escape cancels.
double-click the node itself to toggle it on/off (band 0 and band 3 are off by default, double-click to activate).
right-click any node for the full menu (shape submenu for shelf/bell on outer bands, reset, copy/paste).

low-latency mode

toggle the LIVE button in the plugin UI to switch from standard mode (~93ms latency, full spectral resolution) to LIVE mode (~12ms latency, optimized for tracking and monitoring).

LIVE mode uses a smaller FFT window (1024 samples at 44.1 kHz vs 4096 in standard). the trade-off: slightly less spectral detail in the processing. on most material the difference is subtle. standard mode is better for critical mixing decisions. LIVE mode is better for hearing what the plugin does while you record.

right-click the spectrum display in LIVE mode to choose between LIVE analysis (matches the audio engine's lower resolution) or HQ analysis (runs the full-resolution analyzer alongside the LIVE audio path for a higher-detail visualization).

CPU stays under 3%. only one engine runs at a time. the inactive engine is fully gated. switching between modes uses a 50ms crossfade with no glitch or click.

LIVE mode state is preserved across preset loads and A/B comparisons.

presets

PUSH ships with 36 factory presets, named after the problem they solve rather than the gear they emulate. browse with the < > arrows in the header. compare with the A/B button.

SETTLE presets (15)

preset	use case
Safe Starting Point	neutral default. start here if unsure.
Clean Up Low Mids	removes 200–500 Hz buildup on busy mixes
Tighten Muddy Mix	controls low-mid mud across the entire spectrum
Polish Master Bus	subtle finishing glue, low AMOUNT
Even Out Loud Vocals	holds vocal performance dynamics in check
Glue Loose Drum Bus	SSL-style bus glue for a drum bus that feels separate
Tame Acoustic Peaks	controls acoustic guitar pick attacks and string slaps
Clear Up Synth Stack	thins crowded synth layers in the low-mids
Push Vocal Forward	brings a vocal up in the mix without raising the fader
Tame Harsh Top End	focused control above 4 kHz
Lock Down EDM Bass	tight low-frequency control for sub and bass
New York Bus Blend	moderate AMOUNT with parallel-style MIX, classic NY drum bus character
Pump Pads On Beat	rhythmic ducking on a pad bus (works with sidechain enabled)
Clean Mid-Bass Master	master-bus subtlety, targets 80–250 Hz
Focus Center Vocal	M/S mode, compresses the centre only, leaves stereo content untouched

TIGHT presets (13)

preset	use case
Wake Up Flat Drums	aggressive transient shaping for lifeless drum recordings
Control Loose Kick	fast-attack kick control for trap and EDM
Sharpen Dull Snare	recovers snare snap from a soft hit
Cut Vocal Through Mix	aggressive forward placement for dense mixes
Sharpen Dull Guitars	brings attack and bite back to dull electric guitar takes
Tame Wild Bass	fast control for inconsistent bass performances
Slam The Whole Mix	aggressive mix-bus character, use sparingly
Hit Harder Trap Drums	maximises drum-bus punch in trap and hip-hop
Drive Rock Drums Hard	rock drum bus, classic 1176-style
Maximum Peak Grab	the most aggressive preset, for emergency peak control
NY Parallel Crush	extreme TIGHT, designed for parallel use at MIX 30–50%
Catch Stray Peaks	fast attack for occasional level spikes without affecting the rest
Duck Under Kick	sidechain-style ducking for bass under kick (enable sidechain)

DRIFT presets (8)

preset	use case
Open Up Room Mics	lifts ambient room recordings without amplifying noise floor
Lift Vocal Tails	recovers the breath and sustain at the end of vocal phrases
Reveal Hidden Air	brings out 8–15 kHz top-end sparkle on dull recordings
Open Cymbal Shimmer	extends cymbal decay and recovers HF detail
Wake Up Lifeless Track	broad upward across the spectrum, last-resort revival
Bloom Dull Pads	extends sustain on synthesised pads and held chords
Expose Buried Detail	brings up quiet performance details (finger noise, breath, room)
Sustain Sub Bass	extends sub-bass tails without raising the body

how it works

PUSH processes the signal through a 5-stage pipeline. no neural networks. no machine learning. every algorithm choice is deterministic and traceable to a specific design intention.

STFT analysis + ERB envelope

a 4096-point STFT splits the signal into frequency bins. 40 perceptually-spaced ERB bands (Glasberg & Moore 1990) map those bins to the way the human ear actually groups frequency. per-band magnitude envelopes feed the compressor, not raw bins.

per-band level detection + crest factor

each band tracks its own envelope. SETTLE and DRIFT use RMS, TIGHT uses peak. crest factor (peak-to-RMS ratio) adapts attack and release per band: percussive material gets faster timing, sustained material gets slower timing. the SPEED knob scales both.

per-band gain computer + broadband coupling

each band runs its own gain computer (Giannoulis et al. 2012) with character-specific thresholds, knees, and ratios. then a broadband GR coupling stage blends the band-by-band gain with an energy-weighted overall gain (SETTLE 0.6, TIGHT 0.4, DRIFT 0.15). this is what makes PUSH feel like a compressor, not a dynamic EQ. the bands share a coherent compression envelope.

cepstral gain smoothing

a cepstral smoother (frequency-domain low-pass on the log-magnitude gain curve) prevents adjacent bands from ringing against each other. character-dependent cutoff: SETTLE 10 bins, TIGHT 12 bins, DRIFT 8 bins. floor blend at 25–30% so the spectrum stays musical at deep GR.

character coloring + auto-compensation

optional harmonic coloring applies per-character: SETTLE adds gentle even harmonics, TIGHT adds a dynamic odd/even blend, DRIFT stays clean. K-weighted loudness compensation (ITU-R BS.1770) matches output to input level. an ISTFT reconstructs the time-domain signal, the dry path is mixed in at the MIX setting, and M/S decoding runs if needed.

result: per-band sensitivity. broadband compression feel. no neural model. every decision is documented and debuggable.

technical specifications

spec	value
formats	VST3, AU (AUv3), standalone
platform	macOS Universal Binary (Intel + Apple Silicon), Windows 10 + native
sample rates	44.1 – 192 kHz
latency	~93 ms at 44.1 kHz (4096 samples, STFT stage)
CPU target	under 3% on macOS at 44.1 kHz, single instance (hardware-accelerated FFT)
frequency bands	40 ERB-spaced (Glasberg & Moore 1990)
characters	3 (SETTLE, TIGHT, DRIFT) with character-specific gain computers, transient gates, releases, and coloring
presets	36 factory presets (15 SETTLE, 13 TIGHT, 8 DRIFT)
sidechain	external stereo or mono sidechain input, routed through the same 40-band detector
auto-compensation	K-weighted RMS (ITU-R BS.1770), on by default
DRM	none. no iLok. no account. no activation server.
license	permanent. one-time online activation, cached offline forever.

installation

macOS

download the installer from kernaudio.io/push
copy KERN PUSH.vst3 to ~/Library/Audio/Plug-Ins/VST3/
copy KERN PUSH.component to ~/Library/Audio/Plug-Ins/Components/
open your DAW, scan for new plugins
find "KERN PUSH" under manufacturer "KERN Audio"

Windows

download the installer from kernaudio.io/push
copy KERN PUSH.vst3 to C:\Program Files\Common Files\VST3\
open your DAW, scan for new plugins
find "KERN PUSH" under manufacturer "KERN Audio"

one-time online activation with your license key, then cached offline permanently. no iLok. no account. no subscription.

quick start recipes

"i want bus glue on my drums"

select SETTLE character
load the Glue Loose Drum Bus preset, or set AMOUNT to 30% and SPEED to 55%
aim for 3–5 dB of gain reduction on peaks
MIX stays at 100%. if you want it more transparent, drop MIX to 70%

"my snare needs grip"

select TIGHT character
load Sharpen Dull Snare, or set AMOUNT to 55%, SPEED to 80%
the transient gate keeps the snap intact while pulling the body down
A/B with bypass. AUTO is on, so the snare should feel tighter, not louder

"i want to bring up the air on a quiet vocal"

select DRIFT character
load Lift Vocal Tails, or set AMOUNT to 40%, SPEED to 35%
FOCUS at 4–6 kHz biases the upward gain toward presence and air, not body
use DELTA to verify: you should hear breath and sustain coming up, not the body of the vocal

"i want it before my Pro-C"

load PUSH first in the chain, then your final compressor after it
set PUSH to SETTLE at AMOUNT 25–30%
aim for 3–4 dB of gain reduction on the loudest moments
now your Pro-C (or 1176, or LA-2A) only needs to do 2–3 dB of additional GR. it sounds more like itself

"i want parallel compression"

load PUSH on the source track or bus directly
select TIGHT character
load NY Parallel Crush, or set AMOUNT to 85% (aggressive)
drop MIX to 30–50%
the crushed wet signal blends in underneath the dry. classic New York parallel compression

"i want stereo width without losing the centre"

switch to M/S stereo mode
load Focus Center Vocal, or set SETTLE AMOUNT 30%, SPEED 40%
PUSH now compresses only the centre channel (vocal, kick, snare), leaving the sides (room, stereo reverb) untouched
the result is a tighter centre and a wider apparent stereo image without any width plugin

tips

DRIFT is the lever. if a mix sounds technically correct but emotionally flat, DRIFT pulls up the detail other compressors do not address. air, room, sustain, breath.
AUTO is honest. on by default. when you A/B against bypass, the loudness is matched so you hear the character, not a volume change. only turn it off when you specifically want to hear raw GR action.
FOCUS targets the band. 1 kHz is the default centre. drop FOCUS toward 200 Hz to make PUSH lean on the low end, raise it toward 8 kHz to make PUSH lean on the top. logarithmic scale, behaves like a parametric EQ centre frequency.
DELTA reveals everything. hit DELTA to hear only what PUSH is doing to the signal. for SETTLE and TIGHT, you hear the removed peaks. for DRIFT, you hear the added detail.
AMOUNT 30 is the home position. most factory presets sit between 20% and 50%. 75%+ is aggressive territory (Slam The Whole Mix, NY Parallel Crush). 15–25% is finishing territory (Polish Master Bus, Clean Mid-Bass Master).
it is the step before, not the step instead. PUSH is designed to sit before your final compressor in the chain, not replace it. 3 dB of GR through PUSH means your Pro-C, 1176, or LA-2A only has to do 2–3 dB more. each plugin does the part it is best at.
M/S mode is transparent. when you want compression on the centre (vocal, kick, snare) without affecting the sides (room, reverb), switch to M/S and compress only what needs it.
browse problem-first. the 36 factory presets are named after problems ("Tame Harsh Top End", "Open Up Room Mics") not gear emulations. find the preset that matches your problem, then adjust AMOUNT and SPEED from there.

troubleshooting

problem	solution
"it feels like dynamic EQ, not compression"	enable AUTO (on by default), and increase AMOUNT past 40%. the broadband GR coupling stage makes the per-band gains feel like a coherent compression envelope rather than 40 independent EQ moves. at very low AMOUNT, the coupling is reduced.
"DRIFT does not do anything at low AMOUNT"	this is by design. DRIFT zero-gains at 0% AMOUNT (silent DELTA). try AMOUNT 30%+ to hear the upward compression. the Reveal Hidden Air preset is a good starting point.
"CPU is much higher than 3%"	"under 3% on macOS at 44.1 kHz, single instance" is the floor. at 96 kHz, expect ~2.2x. at 192 kHz, expect ~4x. on Windows, expect higher CPU than macOS (no hardware-accelerated FFT). older Windows hardware with Asio4All can push CPU well over 30%. drop the sample rate, switch to a native ASIO driver, or use fewer instances.
"the spectrum display 'overshoots' the +12 dB ceiling"	this is a display ceiling, not a DSP overshoot. the visualisation caps at +12 dB. if your source hits 0 dBFS, the per-band magnitude saturates the visible rail. AUTO matches output loudness to input loudness, but it does not change the displayed input level. trim the source by -6 to -12 dB upstream and the spectrum will sit below the ceiling.
"i hear pumping"	SPEED is too high for the source, or AMOUNT is too high. lower SPEED to slow the release. or raise AMOUNT so the GR sits in a more consistent range (counter-intuitive but works on busy material).
"i hear clicks when switching characters"	this should not happen. PUSH crossfades between characters over ~10 ms. if you hear clicks, try restarting the plugin or reloading the project.
"plugin does not appear in DAW"	rescan plugins in your DAW. macOS: check VST3 at `~/Library/Audio/Plug-Ins/VST3/` and AU at `~/Library/Audio/Plug-Ins/Components/`. Windows: check VST3 at `C:\Program Files\Common Files\VST3\`. find under manufacturer "KERN Audio".
"sidechain input has no effect"	SIDECHAIN toggle must be enabled. then route an auxiliary signal to the plugin's sidechain input in your DAW. supports stereo and mono sidechain. the sidechain runs through the same 40-band ERB detector.

keyboard / mouse reference

action	control
adjust knob	click + drag vertically
fine adjust knob	shift + drag
reset knob to default	double-click knob
toggle EQ node on/off	double-click node
type exact EQ values	double-click label text ("250 +0.0 dB Q 1.0")
cycle EQ shape (band 0 or 3)	click type badge (BELL / LS / HS)
switch character	click SETTLE / TIGHT / DRIFT pills
switch stereo mode	click L/R or M/S pill
browse presets	click < > arrows in header
open preset menu	click the preset name between the arrows
toggle A/B comparison	click A/B button
toggle delta mode	click DELTA button
toggle auto-compensation	click AUTO button
toggle sidechain	click SIDECHAIN button
bypass plugin	click power button in header

version history

v1.1.0 (may 28, 2026)

LIVE mode (~12ms latency). all three characters (SETTLE, TIGHT, DRIFT) verified in low-latency. preset recall fix for M/S routing. host-bypass latency reporting. shelf chevron on outer EQ nodes.

v1.0.5 (may 12, 2026) / initial public release

three compression characters: SETTLE (bus glue), TIGHT (transient grip), DRIFT (per-band upward compression)
40 perceptually-spaced ERB bands (Glasberg & Moore 1990) with per-band level detection, gain computer, and release
broadband gain-reduction coupling (ADR-029) so the per-band gains feel like coherent compression rather than dynamic EQ
character-specific timing models: SETTLE dual-time-constant auto-release (SSL bus comp DNA), TIGHT GR-dependent release + program-dependent ratio (1176 DNA), DRIFT history-dependent release (LA-2A T4 thermal-cell DNA)
per-band transient gate with character-specific reduction (SETTLE 30%, TIGHT 85%, DRIFT 0%)
optional character coloring: SETTLE asymmetric tanh (gentle even harmonics, 0.1–0.5% THD), TIGHT dynamic odd/even blend (1–3% THD, 2x oversampling), DRIFT clean
cepstral gain smoothing with character-dependent cutoff (SETTLE 10, TIGHT 12, DRIFT 8)
K-weighted auto-loudness compensation (ITU-R BS.1770), on by default
external sidechain input (stereo or mono), routed through the same 40-band ERB detector
spectral display with pre-curve ghost (dim input behind bright output), peak hold line, transient flash overlay
36 factory presets named after the problem they solve (15 SETTLE, 13 TIGHT, 8 DRIFT)
under 3% CPU on macOS at 44.1 kHz, single instance. 88 automated tests covering DSP correctness and regression.

KERN PUSH / user manual