the harshness you stacked yourself

a producer sends me a vocal bus and asks which eq move kills the harshness. eight takes of the same hook, stacked for thickness. each take sounded full on its own, he says, but together they have a hard edge that sits at the front of the mix and will not leave. he has already tried a wide dip around 3 kHz. it helped for a second, then the edge came back, somewhere slightly different. i tell him the eq is not the problem, and neither is any single take. the harshness is something he built by stacking, and you cannot subtract your way out of it.

the frame

here is the thing nobody mentions when they tell you to “layer it up for thickness.” stacking is not just adding volume. it is adding correlation.

when you sum two signals, what happens depends on how alike they are. two copies of the same sound, lined up in time, add coherently: at the frequencies they share, the level climbs toward six decibels louder, double the pressure. two unrelated sounds add incoherently, around three decibels on average, with no real reinforcement. this is settled physics, the kind Julius Smith at Stanford writes down in one clean line.

now think about what your eight takes actually are. eight versions of the same voice, singing the same notes, in the same room, into the same mic. the things they share, the resonances of that voice, that room, that mic, are correlated across every layer. so those few frequencies reinforce hard, toward the six-decibel end. everything unique to each take, the human wobble, the noise, the microvariation, only adds the gentle three.

stacking does not lift the sound evenly. it takes the handful of frequencies every layer has in common and shoves them forward, while leaving the rest behind. you reached for “bigger” and the physics handed you “narrower and harder” instead.

why it turns into an edge

a tall narrow peak is not automatically harsh. what makes it bite is a second mechanism sitting on top of the first.

back in 1965, two researchers named Plomp and Levelt measured something the ear has always known. when two tones land within about one critical band of each other, close but not identical, you do not hear two clean notes. you hear roughness. the partials beat against each other faster than your ear can separate them, and the sensation is a grating edge. it is worst when they sit about a quarter of a critical band apart.

your stack is full of these. eight takes are never perfectly in tune with each other, so the shared resonance at 3 kHz is not one frequency, it is a little cloud of near-misses, all reinforced, all sitting inside the same critical band, all beating against each other. that is the edge. it is not loudness, it is roughness, and you manufactured it by piling correlated, slightly detuned copies into the same narrow space. William Sethares modelled this cleanly in the nineties: the total roughness of a sound is the sum of every rough little pair inside it. stack more, sum more pairs, get more edge.

this is not only a studio problem

i think about this one a lot outside the mix, because it is the cleanest physical model i know for something that is true everywhere.

repeat one move enough times and you do not amplify the whole of it. you amplify the part that is the same every time. ten marketing emails with the same hook do not land ten times harder, they make the one tic in your writing impossible to ignore. run the same optimisation on your calendar every week and you do not get ten times the output, you get the one flaw in the system reinforced until it is the only thing you feel. correlation compounds. the shared frequency wins. it is the same math whether it is a vocal stack or a habit.

why an eq cannot save you

so why does the wide dip fail. two reasons. the buildup is dynamic, it only spikes when the layers happen to align, so a static notch is wrong most of the time, cutting a hole when there is no problem or missing the moment when there is. and the resonance is narrow and moving, sitting in a different near-miss from phrase to phrase, so a fixed band is always slightly in the wrong place.

what works is to catch the reinforced peaks only when they reinforce, and only in the band where they live. per frequency band, dynamically, weighted the way the ear weights it. that is the whole job SMOOTH does, and it is why i built it on forty bands spaced the way the cochlea is spaced (the ERB scale, from Glasberg and Moore’s 1990 work) rather than evenly. evenly spaced bands waste resolution where your ear has none and starve it where your ear is sharpest. but honestly, the tool is the second half of the lesson. the first half is just knowing what you stacked.

what this changes

next time a stack turns brittle, do not start subtracting from the bus. start one layer back. thin the shared resonance a little on each take before you stack, so there is less correlated edge to reinforce in the first place. then, if it still bites, use a dynamic per-band suppressor on the bus to catch the peaks when they spike, instead of a static eq to flatten them always.

and a smaller, free move: before you commit a big stack, solo it and listen for the edge on its own. if it is rough alone, it will be rougher in the mix. you stacked it. you can un-stack a little.

from the studio

this one came from my own sessions as much as the inbox. i have a folder of vocal stacks from my artist and producer days that taught me this the expensive way, by sounding huge in headphones and brittle on a club system.

what do you reach for when a layered part turns harsh? an eq, a de-esser, fewer layers, something else? reply, i read all of them, and the good corrections become the next one.

jonas