the 3 kHz you hear is not the 3 kHz they hear

you cut something around 3 kHz on almost every mix. a couple of dB, sometimes more, usually with a narrow band because it felt surgical and responsible. and you’ve never been fully sure you got it right, because the next day it sounds like you took too much, and on someone else’s system it sounds like you took nothing at all.

that doubt is the correct response to a real problem, and the problem isn’t in your monitors.

the frame

your judgments below about 1 kHz transfer to other listeners. your judgments between 2 and 5 kHz don’t, because the outer ear has a resonance sitting in that region and it varies between normal, healthy ears by roughly 2 kHz in frequency and 10 dB in level. you’re not measuring the mix there. you’re measuring the mix through a piece of anatomy nobody else owns.

the part of your hearing that is furniture

before a sound reaches your eardrum it has to get past your torso, your head, the flange of your outer ear, the bowl of the concha, and then down a tube about 25 mm long. none of that is neutral. the canal behaves like a pipe closed at one end, which gives it a quarter-wave resonance at around 2.7 kHz in an adult. the concha has its own, separate mode higher up, nearer 5 kHz. these two get lumped together constantly, including by people who should know better, but they are different mechanisms with different frequencies.

Shaw mapped this in 1974, gathering twelve prior studies into one transform from the free field to the eardrum (J. Acoust. Soc. Am. 56, 1848). the picture hasn’t changed much since. there is a broad lift through the 2 to 5 kHz region, and its size depends on where the sound is coming from.

now, the point where most versions of this story go wrong.

this isn’t “your ear is lying to you”

it is tempting to conclude that the harshness is an artifact your ear invents, and the recording is innocent. that’s wrong, and it’s worth being precise about why.

that outer-ear lift applies to every sound you’ve ever heard. the live drummer in the room went through the same canal. the guitar amp, the voice across the table, the reference track. the lift is the shape of your hearing rather than something your mix does to you, present in all of it, and your brain has spent your whole life accounting for it.

so the resonance is only half the story. the half that matters is that it isn’t the same resonance in the next person’s head.

the number that should bother you

Denk and colleagues measured external-ear acoustics across a population of normal ears (Trends in Hearing, 2018). the main resonance varies in a range of almost 2 kHz in frequency and about 10 dB in level. above the peak, the spread between individuals reaches 20 to 30 dB.

sit with the size of that. two kilohertz of uncertainty about where the emphasis is. ten decibels about how much. and this is across ordinary, healthy hearing, not damage and not age.

your low mids don’t behave like this. the transfer at 200 Hz is close to identical from head to head, because at those wavelengths nobody’s ear is big enough to matter. that’s why your bass decisions travel and feel solid, and it’s why the top of the midrange is the region where mixes fall apart on other systems.

this isn’t a solved problem, either. the current research effort is trying to predict an individual’s high-frequency response from body measurements using machine learning, and it’s still hard: recent work on individualizing head-related transfer functions (Niu, Koyama and Nakamura, arXiv:2508.16176, 2025) exists because you can’t generalize one ear’s behaviour above a few kHz to another’s.

what a narrow cut at 3 kHz actually is

here’s where it gets uncomfortable. how sharp is your hearing at 3 kHz? the auditory filter there is about 350 Hz wide (Glasberg and Moore, Hearing Research 47, 1990). that’s roughly a ninth of an octave. it’s narrow.

so a surgical 1/6-octave notch at 3 kHz lands on the ear as a cut comparable in width to the resolution of your hearing, aimed at a target whose position you don’t know within two kilohertz, based on a reading taken through your own ear canal. nothing subtle about it.

you’re using a scalpel on coordinates you don’t have.

what to do about it

not much, and that’s the point.

cut less than you think. if you’re reaching for 4 dB at 3 kHz, the honest version is probably 2, because some of what you’re hearing is your own anatomy and your listener’s will differ.

cut wider, not narrower. a broad, gentle move degrades gracefully when the target turns out to be 700 Hz away from where you thought. a narrow one either hits nothing or leaves a hole.

prefer something that only acts when the problem is actually there. a static cut is a permanent bet on a frequency you’re unsure of. a dynamic one is only wrong while the resonance is ringing.

and check on a second system, which you already know, but now you know why it’s that band specifically that betrays you.

the same shape outside the studio

you have met this before. you find someone’s voice grating and your friend doesn’t hear it. neither of you is wrong and neither is being difficult. you’re running the same signal through different hardware, and the disagreement lives in a narrow band you both feel certain about. certainty isn’t accuracy. it never was.

the tools

this is the thinking inside SMOOTH. it works in ERB bands, which follow the width of your hearing rather than a convenient grid, so a move is broad where your hearing is broad. and it’s dynamic: it acts on a resonance while the resonance is present and gets out of the way when it isn’t. that’s deliberate. when the target is genuinely uncertain, the cheapest mistake is the one that only lasts as long as the problem does.

from the studio

tell me about the mix that was harsh to you and fine to everyone else, or the 3 kHz cut you undid a week later. i read every reply, and they turn into the next of these.

jonas