Harshmallow FAQ

Harshmallow is an adaptive multiband de-harsher. It watches each band and pulls down harsh upper-mid and treble resonances only while they flare, then releases - so the source keeps its character. It is not a static EQ cut: it reacts and lets go, tuned by ear with four SHAPE controls and two TIME controls per band, across Stereo, independent L/R, or Mid/Side.

Compared with a typical de-esser, Harshmallow is not tied to one fixed sibilance region. Compared with a general dynamic EQ, it is purpose-built as an attenuation-only de-harshing tool, with multiband lanes, Stereo/L/R/M/S routing, Guard/Focus shaping, per-band timing, and a live correction trace showing the reduction.

The de-harshing path is attenuation-only: it pulls detected peaks down and never boosts them. Input gain, output gain, and Auto Gain are separate level-matching tools, independent of the correction path.

On any track or bus where harshness builds up - vocals, drum overheads, guitars, or a full mix. Place it after a de-esser when the de-esser only catches part of the problem, or on a group/mix bus for general glare. It is stereo in / stereo out and handles mono sources automatically.

VST3, CLAP, and Standalone for Windows 10/11 64-bit and Linux (glibc 2.35+: Ubuntu 22.04+, Debian 12+, Mint 21+, Fedora 36+, Arch). 64-bit x86. macOS is not part of the initial release. Harshmallow does not use MIDI.

VST3 goes to C:\Program Files\Common Files\VST3\ on Windows or ~/.vst3/ on Linux; CLAP to C:\Program Files\Common Files\CLAP\ or ~/.clap/. The Standalone runs from the extracted folder. One licence unlocks all three formats.

Harshmallow runs from 44.1 kHz up to 192 kHz. The reported latency depends on the active engine mode and is reported to compatible hosts for delay compensation.

Activation is local: your key is verified on your machine, with no phone-home and no usage tracking - internet is only needed to download or purchase. The licence is tied to your email rather than a single machine, so you can use it on the machines you own. One licence unlocks VST3, CLAP, and Standalone.

The demo runs the full plugin with a 2-second silence interruption every 60 seconds. The trial chip in the header pulses red so you know what happened. Activating with a valid key removes the interruption immediately, from the first audio block.

Yes. The licence is tied to your email, not a dongle or a single machine, so you can activate on the machines you own. If you reinstall or move to a new machine, paste the same key again.

SENS (Sensitivity) sets when a band reacts; DEPTH sets the maximum reduction; GUARD (Character Guard) protects the source's natural tone from over-processing; FOCUS (Sharpness Focus) sets how narrowly the band targets the peak - high for a single piercing tone, low for broad glare. They are per band, so each split is shaped independently.

They are the per-band timing. ATT (Attack, 1-250 ms) sets how fast the reduction engages once a resonance appears. Short ATT catches sibilance and transient spikes quickly; longer ATT lets more brief transient energy through before the reduction fully engages. REL (Release, 10-2000 ms) sets how fast the band lets go - short for transparent recovery on busy material, longer for a smoother result across a phrase.

Each lane splits into LO, MID, and HI with two draggable crossovers and a per-band Mix marker. Out of the box both crossovers sit at the chart edges, so the lane is a single full-range de-harsher; drag a crossover inward (the pill reads 'Create band') to bring a band into play. The half-octave minimum keeps the bands ordered LO < MID < HI.

Each band has its own Mix marker. It scales how much of that band's correction is applied - 100 % is the band's full de-harshing, 0 % applies no reduction and skips that band's de-harshing stage entirely (which also keeps CPU down). Mix is set per band and per lane.

Linked Stereo (the default) treats left and right together in one lane. Independent L/R splits into separate LEFT and RIGHT lanes; Mid/Side splits into MID and SIDE. In the independent modes each lane has its own crossover positions and per-band Mix, while the per-band SHAPE/TIME are shared across the two lanes so the sides stay tonally matched.

Turn LINK L/R off and engage M/S so you have MID and SIDE lanes. Dial the SHAPE/TIME you want, then bypass the MID lane (its power button) or set its band Mix to 0 % so only the SIDE lane applies the correction. Because SHAPE/TIME are shared per band across lanes, gating the MID lane is what keeps the centre clean.

Live for low-latency tracking, Quality/Studio for everyday mixing (the default), and Deep for a higher-resolution mastering-context pass. Each mode reports a fixed latency to the host, increasing from Live to Quality/Studio to Deep - there is no zero-latency mode.

Harshmallow uses per-band Mix markers instead, so you control how much correction each band applies without blurring the whole plugin into one global wet/dry compromise. There is no global dry/wet knob: the main signal stays in the normal processing path, while each band's Mix controls how much of that band's correction is applied.

Leave LINK L/R on, loop the harshest phrase, and raise SENS until the edge tucks in. Add DEPTH only if the reduction is too light, and raise GUARD if the source starts to lose character. Match levels with the OUTPUT gain (or Output Auto Gain) and bypass to confirm you are judging tone, not loudness.

Drag the upper crossover so the HI band starts above the source's body, then set the LO and MID Mix to 0 % so only the HI band acts. Raise SENS and FOCUS on the HI band to concentrate the reduction on the sharp top.

The full plugin state is saved inside your host project and recalled with it. There is a Default factory preset and an INIT button that resets every control in one undoable step. A/B holds two complete settings to compare, and Copy/Paste moves a setting between instances through the clipboard.

Check MASTER BYPASS and the lane power button, and confirm the band you expect to work has SENS, DEPTH, and Mix above 0 %. If you are working on the LO or HI band, make sure its crossover has been dragged inward so the band actually has width. Also make sure the source is crossing the sensitivity threshold.

Lower SENS or DEPTH, or raise GUARD to protect more of the source's character. Raise FOCUS if you want the reduction to concentrate more narrowly on the harsh peak and leave surrounding frequencies alone.

Lengthen REL so the band recovers more smoothly, or lower SENS so it triggers less often. If it is missing fast spikes, shorten ATT so it engages sooner.

A brief interruption can happen when changing engine mode because the reported latency changes and the host may need to re-sync. Pick the mode before a critical pass rather than automating it during playback; stop and restart playback if your host does not handle the latency change cleanly.