Psychoacoustic Audio Tools
You hear things other people can't name. The difference between a mix that works and a record that's undeniable.
BLACKGLASS is eight mixing and mastering plugins built on published auditory research. They work on what the listener's brain does with your signal, and they fix what's actually wrong instead of decorating it.
One tool for each stage of the chain, first channel to final limiter. Pick the problem that bugs you most. Every plugin has a free 14-day trial.
Put it on one mix. You'll feel it in seconds.
Philosophy
Other plugins change your signal. Ours change how your listener's brain reacts to it. Same audio in, different perception out.
Most plugins optimize the signal. Ours optimize the neural response to that signal.
The brain doesn't hear frequencies in isolation. It hears relationships between them. That's a different problem, and it needs different math.
If a parameter can't be traced to a documented neural mechanism, it doesn't exist in our plugins. No "vibe" knobs. No mystery sauce. You deserve to know exactly what's happening to your signal and why it works.
These won't help you write a better song or arrange a better drop. That's your job. Ours starts after yours is done. The gap between "this is good" and "this is a record" is where most tools fall short, and it's the only part we care about.
The Plugins
Each one solves a single problem, and solves it in a way nothing else on the market does: by modeling the listener. Buy only what you need, or take everything on one subscription. Every plugin is a free 14-day trial first. Full version, no card.
Which do I need?
You don't need all eight on day one. Find the thing that's bugging you, reach for the tool that fixes it, and put it where it belongs in the chain.
| If the problem is… | Reach for | It sits… |
|---|---|---|
| Every channel needs the basics, fast | STRIPPER | on the channel |
| A thin, sterile, digital-sounding source | LAZARUS | on the channel |
| A stiff, flat, lifeless lead vocal | HUMAN | on the vocal |
| A cluttered mix that EQ won't fix | DEPTH | across your buses |
| The wrong element grabbing attention | GAZE | across your buses |
| Muddy or dull at the same loudness | STIMULATE | on the mix bus |
| Technically right, but not "a record" | COMPLETENESS | on the master |
| A flat, narrow, or uncompetitive master | FINALIZER | last, on the master |
Rough order down the chain: STRIPPER · LAZARUS · HUMAN on the way in → DEPTH · GAZE across the session → STIMULATE on the mix → COMPLETENESS, then FINALIZER to finish.
Not sure? FINALIZER covers the whole master bus on its own, and every plugin is free for 14 days. Start with whatever bugs you most.
COMPLETENESS
Your mix is done. The balance is there, the levels are there, everything is where it should be. But it doesn't sound finished. There's a gap between your master and a record, and nothing in your chain closes it. EQ only reshapes what's already there. Saturation stacks harmonics on top of harmonics on top of other harmonics until your mix crowds itself toward white noise. Louder, not more finished.
COMPLETENESS goes last. After your EQ, after your compression, after everything you've done. Right before your final true peak limiter. It analyzes what's psychoacoustically absent from your signal and constructs it. Surgically, with almost no CPU overhead, in a way that's entirely transparent and musical. No harmonics added, no saturation. No reshaping. Actual gap-filling at the spectral level.
You have to hear it to believe it. Put it on one master.
1
Insert
Master bus
Placement
2
Knobs to adjust
0dB
Level change
COMPLETENESS doesn't correct problems. It identifies the psychoacoustic properties present in finished-sounding records that your mix is missing, and adds them back. It's the difference between fixing something broken and completing something unfinished.
Automatic gain compensation on A/B comparisons. You hear the actual processing, not a level difference. When you flip between dry and wet and the mix sounds more like a record, that's the plugin. Not your volume knob lying to you.
Which brings us to…
DEPTH
You know the feeling. Every channel is clean, the frequency balance checks out, the levels are right. But the mix still sounds congested and flat. You reach for another EQ, another compressor. None of it helps, because you're fixing channels when the problem is between them. Dozens of sources all fighting for your listener's attention at once. DEPTH works on that.
All
Instances linked
0
Setup steps
2
Controls
60+
Cited papers
Every instance of DEPTH talks to every other instance in real time. That's how it sees relationships between channels. But cross-instance communication isn't the magic. The magic is what it does with that picture: sub-perceptual shifts in spectral behavior that make the vocal pop out, the bass fuse into one solid foundation, the stereo field open up, and the background elements become a stage instead of a fight club.
All without touching a single EQ curve.
Each element gets continuously ranked on how much spectral movement it has, how transient-dense, how much presence-band energy. Background elements get gently calmed so your ear reads them as space instead of competing events. The sound is fully intact. The brain just places it differently.
No frequency cuts. No dynamics processing. No stereo manipulation. No phase or time tricks. Gain-domain only. The width and warmth you hear after inserting DEPTH is perceptual. Your frequency balance stays exactly where you put it.
Insert on each bus. Instances auto-connect. Adjust Separation and Priority to taste. That's it.
Most people don't hear it at first. They listen for a while, then they bypass, and then they go "okay, wow." That's the effect. You don't notice what DEPTH adds. You notice what disappears when it's gone. That's how sub-perceptual processing works, and it's backed by six decades of published neuroscience on how the auditory system decides what matters and what fades into the background.
It's hard to explain, but here's an attempt: With any sound source, your brain works to discern what's important (signal), and what's background (the stage). In most mixes, elements compete for attention; the pad you have in the background, or the sub frequencies in your mix, even if EQ'd correctly, may be doing things on a psychoacoustic level that screams "pay attention to me!" to your brain. DEPTH helps your brain discern what's important, and what isn't. We do this by measuring more than 28 individual diagnostics for how your auditory circuit perceives "importance", and then we gently nudge the elements in the correct direction to help your brain process, and enjoy, the music. Instead of working to figure out what's going on.
The perceptual result most professionals notice is a type of "glue" that just feels really good. To untrained ears, most listeners will tell you they prefer the corrected version but they won't be able to articulate why. One thing we kept hearing when blind-testing the plugin was "I just feel that version deeper" or "I really enjoy listening to this one".
When it comes to mix engineers, some hear it right away, others need a second. All the individual changes are below perceptual threshold. This is by design. We're not trying to change your mix, that's yours to express however you want. What's common for everyone who's been testing our algorithm is a "wow, okay, that's really wild" when we bypassed the engine after having gotten used to the corrected version.
The result, scientifically speaking, is music that's easier for the brain to process. Think of the difference between following a conversation in a crowded room full of people: you clearly identify and easily process the person in front of you, while everyone else merges into "background"/"stage". Now imagine everyone in the room turning towards you and speaking directly at you. Suddenly your brain is hard at work.
That's a fundamental issue in lots of modern music, and no one's addressed it yet. Quite frankly, we understand why. Decoding the science and turning it into a mathematical model over 200 lines deep took a while. Making it algorithmically heavy but zero-latency and light on your CPU took just as long.
That's DEPTH.
And when you need to know exactly who's winning…
GAZE
Your listener's auditory system re-decides where attention goes every few hundred milliseconds. That allocation is measurable, and your mix either wins it where it matters or loses it to a hi-hat. GAZE runs each element through an auditory front end and scores its salience the way the brain does: audibility under masking, spectral movement, roughness, contrast, sharpness. One dot marks what the ear locks onto right now. When something out-pulls your lead, GAZE names the culprit, tells you which feature stole the ear, and Resolve nudges attention back where you pointed it. Bounded, surgical, bit-exact at zero.
40
ERB bands analyzed
5
Salience features
±6dB
Max correction
0
Change at Resolve 0%
Every GAZE instance publishes its element's salience to a session-wide bus, and each one computes the resulting attention budget: the cocktail-party allocation across everything you've tagged. The dot marks what the ear is actually following. It is rarely the loudest channel.
When a non-lead out-pulls your lead for more than a tenth of a second, that's a collision. GAZE pins it on the culprit's most anomalous feature, so you know whether it stole attention with roughness, flux or brightness. When your lead goes flat in a busy section, that's a void. Both show up before your A&R hears them.
Analysis is half the job. Resolve does the rest: a bounded attention-gain ride, a Texture control for the 30 to 150 Hz roughness the brainstem flags as urgent, and a spectral-contrast nudge. Everything has lookahead, everything is capped, and every move traces back to the diagnosis. At zero, GAZE passes bit-exact and just watches.
Chronic masking gets the scalpel: exact partial-loudness accounting per band answers "how much audibility is this element taking from the lead," and persistent theft engages up to five frozen dynamic bells, capped at 3 dB each. Simultaneous culprits split one responsibility-weighted correction instead of all ducking at once. The Mix Doctor turns the running diagnosis into ranked one-click fixes.
Now, about everything in your mix that's just… standing there.
STIMULATE
In 2005, Dean, Harper and McAlpine showed that neurons in the auditory midbrain adapt to the recent statistics of a sound: input that holds still quickly stops driving them. The listener no longer hears it as an event. But the energy is still physically there, and it still raises the masked threshold across its critical band, so everything alive above it needs more level to read. That is what "muddy" is. Pad tails, ring-outs, reverb wash, the stationary body of a sustained synth. Mass without information.
STIMULATE measures how every region of the spectrum behaves over time, treats sustained stillness as mass, and clears it. It can also reinforce what moves. The lift is capped at whatever the cut removed, so an A/B can never flatter you with loudness. Press delta and listen to what left. If it sounds like nothing you would miss, that was the mud.
2
Knobs that matter
Δ
Hear what leaves
3
Analysis resolutions
0
EQ curves touched
Each region of the spectrum is scored on how it changes over time, at three separate resolutions, because the ear wants fine pitch detail in the lows and fine timing in the highs. Loudness never enters the decision. A quiet static pad gets cleared while a loud moving vocal is left alone.
An internal energy budget caps the boost at whatever the cut removed, enforced every frame. STIMULATE cannot function as a loudness maximizer, which means the clarity you hear in a comparison is real and survives level matching.
One button plays only what STIMULATE removed, latency-matched against the dry signal. The best argument for the plugin is thirty seconds spent listening to the sludge it took out of a mix you thought was clean.
And when it's time to call it finished:
FINALIZER FLAGSHIP
Acousticians spent decades measuring why some concert halls always rate above others. One of the strongest predictors turned out to be interaural cross-correlation: how similar the sound at your left ear is to the sound at your right. Beranek's hall surveys and Ando's preference studies put the best rooms in a specific low-correlation zone, in specific frequency bands. The ear reads that zone as space.
FINALIZER's SPACE engine moves a master into that zone. Only on sustained content, only in the bands where the ear measures spaciousness, and constructed so that a mono fold-down removes the ingredient completely. No cancellation, no comb filtering, no gamble on the club PA.
SPACE is one of six engines, and every one of them is built on a different published perceptual mechanism: the 1/f statistics of natural sound, the ear's adaptation to sustained energy, the threshold below which change is felt rather than heard, masking-model gap filling, the grouping cues that fuse many parts into one performance. Each is described below, by name. K-weighted gain matching holds the whole chain loudness-honest to within 0.02 LU, and a true-peak limiter holds the ceiling.
One insert. About one percent of one core.
6
Engines, one insert
≈1%
Of one core
≤0.02LU
A/B honesty
0
Mono casualties
Conventional wideners decorrelate the channels and pay for it in mono, where the effect collapses into comb filtering. SPACE injects the low correlation the ear reads as a good room, in a form a mono sum removes cleanly. Measured on real masters: correlation tracks the knob, and the mono fold stays identical to the input. Even dead-mono sources develop real width.
Voss and Clarke showed in 1975 that music shares 1/f structure with natural sound, in its spectrum and in its loudness fluctuations. Limiting flattens the fluctuations. Dense processing tilts the slope. THAW measures both numbers on your master and walks them back toward the shape the ear evolved listening to.
Auditory neurons reduce their response to sustained content (Dean, Harper and McAlpine, 2005), so a static tonal balance reads duller in the third chorus than it did in the first. GRIP tracks that adaptation per band and offsets it, zero-sum. Tone stays vivid and loudness never creeps.
Classic psychophysics puts the just-noticeable difference for level near half a decibel. PLEASURE keeps every band in continuous micro-motion at two hundredths of a decibel, roughly twenty-five times below that threshold. Nothing ever registers as change, but the ear's movement detectors never get to file the master as static.
The gap-filling engine from COMPLETENESS, run under a masking model so it generates energy only where the ear registers absence. A second mode, REACH, extends band-limited masters past their spectral cutoff. Nothing that already exists gets boosted or reshaped.
Bodies never repeat a hit exactly, and the ear files exact repetition as mechanical. HAND gives repeats micro-character and leans in with the program the way an engineer rides a fader. Its floor breathes with the music, which matters: comodulation, everything moving together, is one of the ear's strongest cues for hearing a single performance (Hall, Haggard and Fernandes, 1984).
Of course, records are made of performances.
HUMAN
Clinical voice science has numbers for what a living voice does. Healthy sustained phonation carries a fraction of a percent of cycle-to-cycle pitch variation and a few percent of amplitude variation. Breath tracks effort. Pitch drifts through a phrase and re-centers in the gaps. Comping, tuning and editing pull a vocal toward zero on all of those measurements, and the ear registers the loss long before the mind can name it.
HUMAN puts the measurements back. Its movement comes from a motor-control model, bounded correlated wander rather than random wobble, and the performance itself drives it: scoops into onsets after a rest, drift inside phrases, vibrato arriving late on sustains, breath riding consonants and phrase tails. One shared effort signal couples everything, so pushing means slightly sharper, brighter, breathier and closer to the mic, all at once. That coupling is how throats work. It is also the part a randomizer can never fake.
7
Movement systems
1
Shared effort signal
±cents
Not semitones
12ms
Latency, compensated
HUMAN reads phrases, onsets and sustains in the incoming vocal and times its movement to them. Scoops land on entrances after a gap. Drift accumulates inside a phrase and settles between phrases. Vibrato waits until a note has actually been held for a while, the way singers do it.
When the vocal pushes, pitch rises a hair, the top opens, breath increases, and the voice moves in on the mic, together, as one coupled system. Independent randomizers can approximate each of those. None of them can couple them.
Each instance gets its own random character, saved with the session, so a recall renders identically to the bounce. Doubles and harmonies each move on their own, which is a large part of why stacked real singers sound expensive.
Set it, listen for a minute, bypass it. The dry version now reads noticeably flat. Most of HUMAN's work sits below the level of conscious attention, which is where liveness lives.
Some tracks don't need a psychologist. They need a defibrillator.
LAZARUS
The ear decides what belongs to a sound by harmonicity. Partials that land on integer multiples of the fundamental fuse into the note, which is why harmonic distortion can read as warmth and density instead of damage. Partials that land anywhere else split off and read as a second, uglier sound riding on top. Cheap saturation aliases: it folds the harmonics it creates back across the spectrum onto non-integer frequencies, and the ear files the result under digital fizz.
LAZARUS runs its two stages at up to 768 kHz internally, so everything it adds lands exactly on the harmonic series. The even-order stage brings weight and body. The odd-order stage brings presence and bite. Series or parallel, each dialed from a whisper to a shove, repeatable to the decibel.
It costs $19 because simple things done properly should cost simple money.
2
Harmonic stages
768k
Max internal rate
<2ms
Latency, compensated
$19
Seriously
Two independent stages with different personalities. The even stage thickens: weight, warmth, that through-a-good-console solidity. The odd stage sharpens: presence and attitude. Use one, use both, decide how they meet.
At up to 768 kHz internal, the alias products land around a hundred decibels down, below anything a listener will ever meet. Crank the drive on a bright synth and watch the analyzer. The added partials sit on the harmonic series and nowhere else.
Series stacks the stages for compounding density. Parallel blends them for a more open enhancement. Both voicings are calibrated: the amount you set is the amount you get, session after session.
And finally, the fun one.
STRIPPER
Not every problem needs a neuroscience paper. Sometimes a channel needs the basics, done fast, sounding good: clean gain, a musical EQ that turns surgical when you ask, mid/side when you want it, and a spectrum analyzer we polished far past what a $19 plugin deserves. Smooth, quick, peak-holding, tilted so balanced music reads flat. There is also a little "sweeten" on the way out, for when a track needs to be five percent prettier and you don't want to think about why.
STRIPPER is our fun extra. Light enough for every channel, quick enough that you never break flow, and the only BLACKGLASS plugin with zero citations.
M/S
When you want it
60fps
Analyzer
Every
Channel, no guilt
0
Citations
Fluid at 60 fps, fractional-octave smoothed, with a peak-hold trace and interpolated lows that don't collapse into a staircase. You'll catch yourself opening STRIPPER just to look at things.
Broad, forgiving curves for everyday shaping. Narrow precision and a linear-phase mode when something actually needs cutting out. The sweeten stage adds a gentle finishing character for when correct isn't quite nice.
A channel strip you can instantiate forty times without thinking about your CPU or your session's mood. Open, tweak, close, next.
The Science
Auditory Scene Analysis
Elements with rapid spectral change grab attention. Steady-state signals get filed as environment. Bregman's work on auditory streaming, combined with decades of masking research, is the foundation of how DEPTH decides what's foreground and what's space in your mix.
Bregman, A.S. (1990), Auditory Scene Analysis, MIT Press
Perceptual Masking
When too many sources occupy the same attentional bandwidth, the ear can't separate them. The congestion you hear in a dense mix isn't about overlapping frequencies. It's about overlapping demands on your listener's attention. Moore's psychoacoustic masking models describe the mechanism. DEPTH addresses it.
Moore, B.C.J. (2012), An Introduction to the Psychology of Hearing
1/f Power Law
Voss & Clarke showed that virtually all music and natural acoustic environments have 1/f spectral structure. The auditory system is tuned for that shape. When processing flattens it, things start sounding synthetic. COMPLETENESS measures what's missing from your signal's statistical profile and restores it.
Voss, R.F. & Clarke, J. (1975), Nature
Neural Adaptation
Neurons in the ventral cochlear nucleus suppress their response to spectrally stationary input. Your brain literally tunes out energy that isn't moving. This mechanism sits at the center of DEPTH, COMPLETENESS, STIMULATE and FINALIZER's adaptation engine.
Dean, Harper & McAlpine (2005), J. Neurosci.
Room Acoustics
Rank the world's concert halls by listener preference and one measurement keeps predicting the order: interaural cross-correlation, how alike the left-ear and right-ear signals are. The best rooms sit in a low-correlation zone in specific bands, which the ear reads as spaciousness. FINALIZER's width engine targets that zone directly.
Hidaka, Beranek & Okano (1995), JASA · Ando (1985), Concert Hall Acoustics
Voice Science
Clinical voice research quantifies healthy phonation: fractions of a percent of cycle-to-cycle pitch perturbation, a few percent of amplitude perturbation, all of it varying with effort. Production tools pull vocals toward zero on every measure. HUMAN restores the statistics of a body holding a note.
Teixeira, Oliveira & Lopes (2013), Procedia Technology
We start with a documented phenomenon in auditory neuroscience. A measurable neural response with known parameters and frequency behavior. Not "warmth" or "presence" in the hand-wavy sense. The real thing.
The research dictates the architecture. Where to process, what frequency ranges matter, which neural thresholds to target. If the literature says the modulation filterbank peaks at 4–6 Hz with a Q around 1, that's where the algorithm operates. The science decides what the plugin does.
The science decides what. The defaults decide how much. We calibrate across hundreds of sessions with Grammy-nominated producers and mix engineers whose work sits across billions of streams. They tell us where the sweet spots land on real material, not in theory.
Pricing
Every plugin comes with a 14-day trial. Full version, no feature gates, no credit card. Buy a plugin once and it's yours forever, or get every plugin we make on one $19/month subscription and cancel whenever. Every license covers two machines.
Prices in USD · VAT/sales tax handled at checkout by Stripe · Secure card, Apple Pay & Google Pay
2 machine activations per license · 30-day offline grace · 14-day money-back guarantee on purchases
Downloads
Trial and full versions are the same download. Install, open the plugin in your DAW, paste your license key. Activation takes seconds and works for 30 days between online checks, so a session never stops because your studio machine is offline.
FAQ
Enter your email, get a license key, download the installer. The trial is the full plugin for 14 days: every feature, no watermarks, no interruptions. When it expires, the plugin stops processing until you add a paid key. Your DAW sessions and settings survive.
Start with the problem, not the plugin. The Which do I need? guide maps common problems (boxy tracks, cluttered mixes, flat masters) to the tool that fixes each one and shows where it goes in your chain. Then trial that one for free. You don't need all eight on day one.
Buying is a perpetual license: pay once, use that plugin forever, including all updates within the major version. The subscription is $19/month for every plugin we make, all eight today and everything we ship next, cancel anytime. If you cancel, the plugins keep working until the end of your paid month. Subscribe to try the whole toolbox or cover a busy stretch; buy the one or two you reach for every day.
Two at a time. Studio rig and laptop, for instance. You can deactivate a machine from inside the plugin to free the seat for another. If a machine dies before you could deactivate, contact us and we'll release the seat.
Only for activation. After that, the license lives on your machine and revalidates quietly in the background. You get 30 days of fully offline use between checks, enough for any studio lockout, tour, or cabin mix retreat.
AU and VST3 on macOS (Intel and Apple Silicon, notarized installers). Windows VST3 builds are in development. Write to us and we'll tell you the moment they land.
14 days, no interrogation, on purchases. If it doesn't do for your mixes what this page says it does, reply to your receipt email and we'll refund it. Subscriptions can be cancelled anytime but aren't refunded pro-rata. Full details in the refund policy.
Use the recovery form in the support section. Enter the email you bought with and every active key you own lands in your inbox.
No. DEPTH is gain-domain only (no EQ curves, no dynamics, no phase tricks), driven by a model of auditory attention. COMPLETENESS generates signal where your spectrum has perceptual gaps instead of boosting or distorting what's there. The science section cites the actual papers, and the trial costs nothing.
Support
Enter the email you purchased or trialed with. Every active key registered to it will be re-sent within a minute.
Activation trouble, seat resets, refunds, Windows timelines, or an opinion about what we should build next. It all goes to the same inbox, and we read everything.
Email support