what SMOOTH does
SMOOTH detects and suppresses problematic resonances in your audio — the harsh frequencies that make vocals pierce, synths bite, and mixes fatigue. it analyzes your signal across 40 psychoacoustic frequency bands, identifies which peaks are resonances versus musical character, and transparently reduces only the problems.
think of it as an intelligent, frequency-selective compressor that works on thousands of frequency bins simultaneously, 43 times per second.
controls
SMOOTH has 4 knobs, 2 toggle buttons, 4 draggable EQ nodes, and a mode selector. that's it.
knobs
| control | range | default | what it does |
|---|---|---|---|
DEPTH | 0–24 dB | 6 dB | how much resonance is removed. higher = more aggressive suppression. |
SELECT | 0–12 dB | 4 dB | how prominent a resonance must be before SMOOTH catches it. lower = catches more. higher = only the worst offenders. |
ATTACK | 0.5–50 ms | 2 ms | how quickly SMOOTH reacts. fast (1–5 ms) for transients. slow (20–50 ms) for gentle, transparent processing. |
MIX | 0–100% | 100% | wet/dry blend. use 50–80% for parallel processing. |
toggle buttons
| button | what it does |
|---|---|
DELTA | hear what SMOOTH is removing. if you hear music, dial back DEPTH or raise SELECT. if you hear only metallic resonance — you're dialed in. |
INTENSE | 3x depth multiplier for exaggerated processing preview. use to quickly hear what SMOOTH is targeting, then turn off for mixing. orange when active. |
mode selector
| mode | best for |
|---|---|
Resonance | default. spectral resonance suppression via the 5-stage pipeline. works on everything. |
Transient | time-domain transient shaping. controls attack punch and sustain character. best on drums and percussive material. |
M/S toggle
the L/R ↔ M/S button in the bottom bar switches between stereo and mid/side processing:
- L/R (default): left and right channels processed independently.
- M/S: mid and side channels processed independently. allows surgical treatment of the center image vs. the sides.
sensitivity nodes (EQ-style)
the 4 colored nodes on the spectral display let you shape where SMOOTH is most and least aggressive.
how they work
- drag horizontally: move the node's center frequency (20 Hz – 20 kHz)
- drag vertically: adjust sensitivity (-12 to +12 dB). boost (up) = more aggressive processing. cut (down) = less processing.
- mouse wheel on node: adjust Q (bandwidth). narrow Q = surgical. wide Q = broad.
- shift + drag: fine adjustment (0.2x sensitivity)
- double-click: toggle node on/off
- right-click (M/S mode): choose channel routing — stereo, mid only, or side only
node colors
| node | default frequency | color |
|---|---|---|
| band 0 | 250 Hz | teal |
| band 1 | 2500 Hz | gray |
| band 2 | 800 Hz (off) | orange |
| band 3 | 6000 Hz (off) | purple |
bands 2 and 3 are disabled by default. double-click to activate them.
per-band M/S routing
when in M/S mode, each node can be routed independently:
- stereo (default): applies to both mid and side engines
- mid: only affects the center image (vocals, kick, bass)
- side: only affects the stereo field (reverb, room, width)
right-click a node to change its routing. mid-routed nodes show a green "M" badge. side-routed nodes show a purple "S" badge.
spectral display
the real-time spectral display shows:
- cyan curve: your input signal's frequency spectrum
- orange overlay: gain reduction being applied (taller = more reduction)
- colored fill behind spectrum: sensitivity curve shape from your EQ nodes
display interactions
- scroll wheel: zoom frequency range (centers on cursor)
- double-click: reset zoom to full 20 Hz – 20 kHz range
display range
- frequency: 20 Hz – 20 kHz (log scale)
- level: +12 dB to -60 dB
presets
SMOOTH ships with 32 factory presets organized by use case.
resonance mode
| preset | use case |
|---|---|
| Gentle | subtle, transparent processing for sensitive material |
| Vocal | general vocal resonance taming |
| Drum Bus | balanced drum processing |
| Full Mix | safe mix bus treatment |
| Aggressive | heavy-handed resonance removal |
| Mastering | conservative, mix-bus safe |
| Dark Source | tames muddy low-mid buildup |
| Bright Source | controls harsh high-frequency content |
| De-Harsh Vocal | aggressive 2–5 kHz harshness removal |
| Harsh Synth Tamer | tames resonant synth patches |
| Surgical Strike | precise, deep cuts at specific frequencies |
| Nuke It | maximum processing — hear everything SMOOTH can do |
| Low-End Cleanup | room rumble and mud removal (80–400 Hz) |
| Kick & 808 | sub resonance taming, protects fundamental |
| Bass Clarity | boxiness removal (250–500 Hz) |
transient mode
| preset | use case |
|---|---|
| Drum Punch | enhance drum attack |
| Vocal Presence | add vocal clarity |
| Snare Crack | tighten snare attack |
| Subtle Shape | gentle transient control |
| Cymbal Tame | high-frequency harshness on cymbals (4–12 kHz) |
| Perc Click | control clicky percussion attacks (1.5–3.5 kHz) |
| Transient – Destroy | extreme transient removal |
M/S presets
| preset | use case |
|---|---|
| M/S Vocal Center | clean up the center image |
| M/S Side Tamer | reduce side-channel harshness |
| M/S Low Tighten | mono bass cleanup (common mastering technique) |
| M/S Master Glue | gentle M/S mastering — mid presence + side smoothing |
browsing
- use
< >arrows in the header to browse presets - the preset name and index are displayed between the arrows (e.g., "Vocal (2/32)")
- A/B button: compare two different settings instantly
how it works
SMOOTH uses a 5-stage spectral processing pipeline. no neural networks. no black boxes. just physics.
STFT analysis
your audio is windowed into overlapping frames and transformed to the frequency domain via FFT. at 44.1 kHz, this gives 2,049 frequency bins with 10.77 Hz resolution.
ERB envelope estimation
the spectrum is analyzed across 40 psychoacoustic bands (ERB scale). this means SMOOTH is 7x more sensitive to resonances at 1 kHz than at 8 kHz — matching how your ear actually works.
spectral reassignment
sub-bin frequency precision identifies exactly where each resonance sits, and determines whether each frequency component is tonal (resonance candidate) or noise (leave alone).
per-bin gain computation
for each of the 2,049 bins, SMOOTH computes how much to reduce based on: how prominent the resonance is, how tonal it is, how loud the surrounding spectrum is, and your sensitivity node settings. gains are smoothed over time using a zero-delay feedback integrator for tight, artifact-free tracking.
magnitude-only gain application
gains are applied to magnitude only — phase is never touched. this preserves stereo image, transient shape, and harmonic relationships. broadband energy compensation restores the body that resonance removal would otherwise take away.
technical specifications
| spec | value |
|---|---|
| formats | VST3, AU (AUv3), standalone |
| platform | macOS (Apple Silicon native) |
| sample rates | 44.1 – 192 kHz (SFI: adapts automatically) |
| latency | ~93 ms at 44.1 kHz (4096 samples) |
| CPU target | < 3% at 44.1 kHz, 512-sample buffer |
| FFT size | 4096 (44.1/48 kHz), 8192 (88.2/96 kHz), 16384 (176.4/192 kHz) |
| overlap | 75% (hop = FFT/4) |
| ERB bands | 40 (20 Hz – Nyquist) |
| sensitivity nodes | 4 parametric bells per instance |
| max reduction | 24 dB (72 dB with INTENSE mode) |
| phase modification | none — magnitude only |
| DRM | none. no iLok. no account. no activation. |
installation
- download the installer from kernaudio.io
- run the installer — it copies VST3 and AU to your system plugin folders
- open your DAW, scan for new plugins
- find "KERN SMOOTH" under manufacturer: "KERN Audio"
plugin locations
- VST3:
~/Library/Audio/Plug-Ins/VST3/KERN SMOOTH.vst3 - AU:
~/Library/Audio/Plug-Ins/Components/KERN SMOOTH.component
no activation required. no account needed. just install and use.
quick start recipes
"my vocal is too harsh"
- load preset: Vocal or De-Harsh Vocal
- adjust DEPTH to taste (start at 6–8 dB)
- toggle DELTA to hear what's being removed — should sound metallic, not musical
- if needed, drag band 1 node to the harsh frequency (usually 2–5 kHz) and boost it
"my drum bus has ring"
- load preset: Drum Bus
- set ATTACK to 1–2 ms (fast, preserves transients)
- toggle INTENSE to preview the effect, then disable
- use band 0 to target the specific ringing frequency
"my synth is piercing"
- load preset: Harsh Synth Tamer
- increase DEPTH if needed
- band 1 is pre-positioned at 2.5 kHz — drag it to the problem frequency
- tighten Q (mouse wheel) for surgical treatment
"my mix sounds harsh at loud parts"
- load preset: Full Mix or Mastering
- keep MIX at 50–80% for gentle parallel processing
- set SELECT high (8–10 dB) so only the worst resonances are caught
- use DELTA to verify you're only removing problems
"i want tighter low end in M/S"
- switch to M/S mode (click L/R → M/S)
- load preset: M/S Low Tighten
- this targets low-mid mud in the center while leaving sides untouched
tips
- DELTA is your best friend. always check what you're removing. good delta = metallic resonance. bad delta = music.
- INTENSE is for previewing, not mixing. turn it on to hear what SMOOTH targets, then turn it off.
- less is more. a DEPTH of 4–8 dB handles most material. go higher only when you hear specific problems.
- SELECT controls surgical precision. low SELECT = catches everything. high SELECT = only the loudest resonances.
- fast ATTACK preserves transients. for drums and percussive material, keep ATTACK at 1–5 ms.
- slow ATTACK for transparency. for vocals and mix bus, try ATTACK at 10–30 ms.
- use M/S for mastering. clean up the center (vocals, kick) without touching the stereo field.
- sensitivity nodes are optional. SMOOTH works great with all nodes at 0 dB. use them when you want frequency-specific control.
troubleshooting
| problem | solution |
|---|---|
| "i can't hear any difference" | increase DEPTH. try INTENSE mode. lower SELECT. |
| "it sounds duller" | reduce DEPTH. SMOOTH has auto-compensation, but extreme settings can still thin the sound. |
| "transients are smeared" | reduce ATTACK to 1–2 ms. the transient gate preserves attacks, but very slow attack settings override it. |
| "processing is too aggressive" | raise SELECT (catches fewer resonances). reduce DEPTH. try MIX at 60–70%. |
| "i hear artifacts" | reduce DEPTH below 12 dB. at extreme settings, cepstral smoothing is disabled for maximum surgical precision — artifacts may appear. |
| "CPU is too high" | SMOOTH targets <3% at 44.1 kHz. at higher sample rates, CPU increases proportionally. close other instances if needed. |
| "plugin doesn't appear in DAW" | rescan plugins in your DAW. check that VST3/AU files exist in ~/Library/Audio/Plug-Ins/. |
keyboard / mouse reference
| action | control |
|---|---|
| adjust knob | click + drag vertically |
| fine adjust knob | shift + drag |
| reset knob to default | double-click knob |
| move EQ node frequency | drag horizontally |
| adjust EQ node gain | drag vertically |
| adjust EQ node Q | mouse wheel on node |
| fine adjust EQ node | shift + drag |
| toggle EQ node on/off | double-click node |
| change node routing | right-click node (M/S mode) |
| zoom frequency range | scroll wheel on display |
| reset zoom | double-click on display |
| browse presets | click < > arrows |
| toggle A/B comparison | click A/B button |
| toggle delta mode | click DELTA button |
| toggle intense mode | click INTENSE button |
| switch L/R ↔ M/S | click mode button in bottom bar |
version history
v0.9.1 (february 2026) — pre-release
- 5-stage spectral processing pipeline (STFT → ERB → reassignment → ZDF gain → OLA)
- 16 DSP optimizations (OPT-1 through OPT-11, OPT-A/B/C/D/E)
- ZDF envelope detection (Zavalishin 2012)
- auto-compensation (broadband energy restoration)
- sample-rate-independent processing (44.1–192 kHz)
- Apple Accelerate vectorization
- 4 parametric sensitivity nodes with M/S per-band routing
- INTENSE mode (3x depth preview)
- Catmull-Rom spline spectral display
- 32 factory presets
- 84 automated tests
design & development: Jonas Rosbech. algorithm based on published research by Auger & Flandrin (1995), Ephraim & Malah (1984), Griffin & Lim (1984), Moore & Glasberg (1983), Breithaupt et al. (2007), Zavalishin (2012), Schroter et al. (2022).