Mix Tools
GLASSCUT
A de-esser built on spectral level restoration rather than high-frequency compression: it cuts the S's, leaves the vowels alone, and can take the breaths out between phrases.
Free 10-day trial. The download is the whole collection, 22 plug-ins in one installer, so you can try every one; start GLASSCUT's trial from inside the plug-in when you are ready. Downloading does not start it, and trying one plug-in does not start the others.
- AU (macOS)
- VST3
- AAX
- macOS Universal
- Windows 64-bit
- v1.0.0
What it does
The lisp is in the gain stage, not the detector.
A de-esser is not a high-frequency compressor with a good detector. It is a spectral level-restoration problem, and the artefacts that make de-essers sound bad come overwhelmingly from the gain application, not the detection. A perfect detector driving a fixed 6 kHz shelf still produces a lisp.
GLASSCUT shapes the attenuation to match the sibilant's own measured spectral envelope: a bell whose centre, bandwidth and depth track the measured contour, realised with time-domain filters. Because the notch matches the noise's own shape, the /s/ stays an /s/ and simply gets quieter. That is why it holds together at 12 dB where a shelf lisps at 6.
THRESHOLD is a level you set, −80 to 0 dB on the detector's own scale, and the analyser draws it: the lamp, the laser and the optical bench ride at exactly that height, so an S whose peak rises above the laser is an S being cut. Breath Reduction, off until you switch it on, takes the breaths out between phrases, hissing ones included. It decides 80 ms ahead of the audio, so a breath goes from its first moment while the next word's first consonant stays.
Key features
Harmonicity measured in the time domain
The hardest discrimination is /s/ versus a bright vowel: both put strong energy in 4–10 kHz, and level, centroid and band ratios all fail on a soprano through a bright condenser. The physics separates them cleanly: a vowel's HF energy is a glottal pulse train, periodic at F0; an /s/ is turbulent noise, aperiodic.
Forty times cheaper than the obvious approach
Resolving harmonic structure in the FFT would need a 43 ms window, and a short /s/ lasts 40 ms: unusable. Instead the HF envelope is band-passed, rectified, low-passed and decimated by sixteen, giving a ~3 kHz stream on which a 43 ms correlation window is 128 samples. Same measurement, computed from a 10.7 ms slice.
Spectral Contour Matching instead of a shelf
A fixed shelf or crossover removes the vowel's air and the cymbals along with the /s/. FFT bin masking smears in time and goes watery under heavy reduction. SCM shapes the reduction to the measured contour and avoids both.
Dual-rate detection
A fast path (time-domain filterbank, about 1 ms) gives a provisional gain. A slow path (STFT, about 5 ms) arrives to confirm or veto with confidence and frequency. Classification is inherently latent, because at the onset of an /s/ the analysis window is still half full of the preceding vowel. No algorithm beats that.
An S has to be S-shaped
The classifier's verdict is an opinion; where the energy sits is not. An S concentrates it high, so nothing is cut unless it does, whatever the classifier votes: breath, /f/ and /th/ get no S cut. And the cut ends with the S, holding 8 ms and releasing in 5, rather than carrying into the next vowel.
Breath Reduction looks ahead
A breath and a word's first consonant are the same kind of sound for their first tens of milliseconds, and nothing says which until the voice arrives or does not. So, at Normal and Precise quality, it decides 80 ms ahead and can wait. Breath drawn through the teeth reads as an S, and is told from one by the mouth that made it: inhaled air carries a strong band at 1–2.5 kHz that an S does not. Off, it is bit-transparent and adds no latency.
An exact null at zero reduction
Bit-exact, so it is genuinely transparent when it is not working.
The interface
Audio
Audio demos for GLASSCUT are not published yet.
We would rather show you nothing than show you a demo that was not recorded through this plugin.
Measured
Docs/SIBILANCE-PRECISION.md: ten real vocal recordings (eleven for Breath Reduction), judged from the dry audio by a judge that is not the engine · GlassCutTests, 317 checks · stereo · 48 kHz
- Clean vowels losing more than 1 dB of high band
- 0.0 % of 20,241 frames
- How long the cut outlasts an S, median / worst (Threshold −45 dB)
- 12 ms / 38 ms
- Breath Reduction: median energy removed per inhale
- 22.2 dB, over 62 inhales
- Breath Reduction: sung frames losing more than 1 dB
- 8 of 39,475
- Voiced stack against matched-level fricative noise, autocorrelation
- 1.000 against 0.231
- CPU, stereo, 48 kHz: Breath Reduction off / on
- 0.451 % / 0.465 % of one core (budget 0.6 %)
- Latency: Zero Latency / Normal / Precise
- 0 / 3.0 / 6.0 ms, plus 80 ms in Normal and Precise while Breath Reduction is on
- Null at 0 dB reduction
- exact bit-null
- auval
- passes
Figures are produced by this plugin’s own test suite and benchmark, on the hardware and sample rate named above. They are not estimates.
Controls
THE PANELS
- THRESHOLD
- −80 to 0 dB, default −40, on the detector's own scale. The level above which an S is cut, drawn on the analyser as the laser.
- MAX CUT
- 0–24 dB, default 12. The most any S is cut.
- SENSITIVITY
- ±12 dB. How readily something is called sibilant: it moves the classifier's decision boundary without retuning anything else.
- MODE
- Spectral, Split or Wideband, and every blend between: a bell at the S alone, the S with the air above it, or the whole signal evenly.
- SHARPNESS
- 0–100 %, gentle to firm: the knee. An uncertain detection widens it further, so a doubtful S eases in.
- MIX
- 0–100 %.
- INPUT · OUTPUT
- ±12 dB each. INPUT trims ahead of the detector, so it drives detection as an input trim should.
THE FOOTER
- STEREO
- Linked, Dual Mono, Mid/Side or Adaptive, which links the channels as far as their sibilance coincides.
- QUALITY
- Zero Latency, Normal or Precise: 0, 3 or 6 ms of look-ahead. Precise gives the verdict on an S time to land before the S reaches the gain.
- ADAPT
- Fast, Normal or Locked: how quickly the detector adapts to the material. Locked freezes after five seconds, so a quiet intro or a sparse bridge cannot retune it.
- BREATH
- Breath Reduction: removes up to 24 dB of unpitched sound between phrases. Off by default.
- LISTEN
- Hear what is being removed: the input minus the output.
- BYPASS
- The same switch as the power button in the header.
THE HEADER
- Presets · SAVE
- Step through or choose the factory presets; SAVE writes the current state to a preset file.
Technical details
- Formats, macOS
- AU · VST3 · AAX
- Formats, Windows
- VST3 · AAX
- Operating system
- macOS 10.15 or later · Windows 10 or later, 64-bit
- Architecture, macOS
- Universal: Apple Silicon and Intel
- Architecture, Windows
- 64-bit
- Plugin code
- Glct
- Version
- 1.0.0
- Latency
- 0 ms (Zero Latency), 3.0 ms (Normal) or 6.0 ms (Precise), plus 80 ms in Normal and Precise while Breath Reduction is on. Reported for the current Quality, and re-reported to the host when Breath Reduction is switched.
- CPU
- 0.451 % of one core, stereo at 48 kHz; 0.465 % with Breath Reduction on
- Factory presets
- 5
Default — Lead Vocal, Gentle — Preserve Air, Firm — Harsh Source, Dialogue / Podcast (Fast adaptation, Breath Reduction on) and Mastering — Locked (Locked adaptation, Mid/Side so the image is untouched). Five deliberately: a few that each mean something rather than a long list of near-identical starting points.
Worth knowing
- Breath Reduction leaves some breath on purpose: anything shorter than 80 ms, about the last 40 ms before a note, breath that sounds like a voice or is bright enough to be an S, note tails, and the first breath of a take, before it has heard the singer.
- At Zero Latency there is no look-ahead, so Breath Reduction decides as the audio arrives and removes less: a median 7.9 dB per inhale.
- THRESHOLD is a level, not a setting relative to the singer, so on a vocal recorded much hotter or quieter than usual it will want moving from its −40 dB default.
Also in these bundles
COMPLETE COLLECTION
Get GLASSCUT as part of COMPLETE COLLECTION, 22 plugins worth 708 for 199.
199
Save 509
