Every reviewer has had this happen. You write up a headphone, you are specific about what you heard, and someone replies that on their head it does nothing of the kind. Usually the exchange goes nowhere, because both people are telling the truth and neither has a way to show it.

Sean Olive and Dan Clark presented work in 2025 that puts a number on the problem. Measure the same headphone on a group of real listeners and the spread between them is not a rounding error — on some models it is a few dB, and on others it is more than 10 dB in the bass. Crucially, the spread is not random across models. Some headphones are consistent from person to person and some are not, and the difference is a property of the headphone.

That property has a name in circuit terms, and it is the reason I think it can be measured and published like any other spec: acoustic source impedance.

If that is right, it changes what a review can honestly claim. A headphone with low source impedance behaves roughly the same on most heads, which means a subjective description of it is worth something to a stranger, and a PEQ preset built from a coupler measurement will mostly transfer. A headphone with high source impedance does not, and both the subjective description and the preset are worth much less. Right now nobody tells the reader which one they are holding.

1. The headphone is a source with an impedance

Treat the driver, front cavity and pad as a pressure source $P_s(f)$ in series with a complex acoustic source impedance $Z_s(f)$. The listener contributes an ear load $Z_e(f)$ — pinna, concha, canal, and however much the pad is leaking today.

The headphone as an acoustic Thévenin source

The pressure that actually arrives is a divider:

$$ P(f) = P_s(f) \frac{Z_e(f)}{Z_s(f) + Z_e(f)} $$

This is exactly the same maths as amplifier output impedance versus headphone impedance, and the intuition transfers. When $|Z_s| \ll |Z_e|$, the load barely affects the result — a stiff voltage source into a light load. Change $Z_e$ by 20% and the output moves a fraction of a dB. When $|Z_s|$ is comparable to or larger than $|Z_e|$, the divider is doing real work, and the same 20% change in the listener’s ear moves the response by several dB.

Nothing here is new physics. What is missing is anybody actually measuring $Z_s$ per headphone and publishing it.

2. Why this shows up in the bass first

The load a listener presents is roughly the compliance of the trapped volume under the pad, in parallel with whatever path leaks to ambient around the pad. That leak path is resistive and it varies enormously — hair, glasses arms, jaw shape, how the pad has aged, whether you just pushed the cup with your shoulder against a headrest.

Below a couple of hundred hertz the leak dominates $Z_e$, which is why the leak is where the person-to-person spread lives. But how much spread you get for a given range of leaks depends entirely on $Z_s$:

Same range of pad leaks, two source impedances

Both bundles above are the same six leak areas, from a tight seal to a fairly leaky one, normalised at 1 kHz. The blue headphone has a large front volume and therefore a low $|Z_s|$: the six curves stay within about 4.5 dB of each other at 20 Hz. The red headphone has a small front volume and a high $|Z_s|$: the same six leaks spread over roughly 15 dB. Same ears, same range of fits, very different consequences.

That plot is modelled from the divider equation with plausible lumped values, not measured — I am showing the mechanism, not a result. But it is the shape of the effect Olive and Clark found on real subjects, and it explains why the argument in the comments never resolves. On the red headphone, two people genuinely are hearing bass responses 10 dB apart.

3. Measuring $Z_s$ with two known loads

You do not need an impedance probe. Measure the headphone’s response into two loads whose impedances you know, and solve for the source.

  • Load A (sealed): a standard 711-type ear simulator.
  • Load B (perturbed): the same simulator with a calibrated leak or a known extra shunt volume.

From the complex pressures $P_A(f)$ and $P_B(f)$ and the known loads $Z_A(f)$, $Z_B(f)$:

$$ P_s(f)=\frac{P_A Z_A - P_B Z_B}{Z_A - Z_B}, \qquad Z_s(f)=Z_A\left(\frac{P_s}{P_A}-1\right) $$

$Z_A$ comes from the published impedance curve of the coupler. $Z_B$ is $Z_A$ in parallel with a leak element you have characterised — a short slit or tube, modelled as a resistance plus an acoustic mass.

Two practical points decide whether this works. You must keep phase: SPL magnitude alone is not enough to solve for a complex $Z_s$, so the measurement has to be a proper complex transfer function against a reference, which rules out most casual sweep setups. And $Z_A - Z_B$ must be large enough in the band you care about that you are not dividing by noise, which means the perturbation has to be a substantial one, not a token gap.

4. From $Z_s$ to a consistency figure

Once $P_s$ and $Z_s$ are known, you no longer need subjects. Build a population of plausible ear loads and run each one through the divider:

$$ P^{(k)}(f) = P_s(f) \frac{Z_e^{(k)}(f)}{Z_s(f) + Z_e^{(k)}(f)} $$

Variable Range to sweep What it moves
Canal length ±5–10 mm Canal resonance frequency
Canal volume ±20–40% Midrange coupling
Leak area 0–1 mm² Bass rolloff corner
Clamp force ±2 N Pad compliance and seal

Then take the spread across that population, band by band:

  • σdB (20–200 Hz) — leak sensitivity, i.e. how much your bass depends on your seal
  • σdB (2–6 kHz) — placement sensitivity, i.e. how much depends on where the cup sits on your ear
  • Δ90–10 — the gap between the 90th and 10th percentile listener, which is the number a reader actually cares about

The honest limitation is that the answer is only as good as the population model. If real ear loads vary in ways my lumped model does not capture, the ranking could be wrong in detail even if the mechanism is right. Validating a modelled Δ90–10 against a real multi-subject dataset is the step that would turn this from an idea into a metric, and I have not done it.

5. A cheap surrogate: the Leak Susceptibility Index

The full two-load method needs phase-accurate measurement and a characterised leak, which is more than most people have. There is a much cruder version that anyone with a coupler can run:

  1. Measure the sealed response on the 711 coupler.
  2. Insert a small, repeatable leak — a 0.2 mm feeler gauge, 5 mm wide, in the same place every time.
  3. Take the average level change from 50–200 Hz:

$$ \text{LSI} = \text{Avg}(\Delta\text{SPL}_{50\text{–}200\text{ Hz}}) \text{ dB} $$

Lower LSI means the headphone cares less about your seal. This throws away all the phase information and tells you nothing above the bass, but it is repeatable, it needs no modelling, and it captures the band where the disagreement actually happens. If I only get one number published, this is the one I would publish.

Two obvious extensions, same principle:

Metric Method Band What it tells you
Clamp-Force Sensitivity (CFS) Vary clamp ±2 N 100 Hz–1 kHz, 2–6 kHz Pad compliance robustness
Placement Sensitivity (PS) Rotate ±5°, ±10° 3–8 kHz Off-axis concha loading sensitivity

6. Doing it repeatably

The measurement is easy to get wrong in ways that look like a result:

  • Use adjustable micro-valves rather than shims for the leak. A shim changes the clamp geometry at the same time as the leak area, so you cannot tell which one moved the curve.
  • Reseat between every run. Fit variation is the thing being measured, not an error to be minimised. Five runs without reseating tells you about your mic, not the headphone.
  • Recalibrate the coupler mic each session, because you are looking for a few dB and drift of a few dB is possible.
  • For IEMs, sweep insertion depth and canal volume instead of pad leak. Same divider, different dominant variable.

Where I think this goes

I would like to see two numbers in headphone reviews: the frequency response, which we already have, and a consistency figure, which we do not. The second one changes how you should read the first.

It also directly bears on EQ. A coupler-derived PEQ preset assumes your ears load the headphone the way the coupler does. On a low-$Z_s$ headphone that assumption is close enough that the preset transfers and EQ does what it says on the tin. On a high-$Z_s$ headphone the preset was built for a load you do not have, and the bass part of it in particular is a guess — which is a much more useful thing to tell a reader than “results may vary”.

This is a proposal, not a finished method. I have not built the two-load rig, and until I have I cannot rank anything. What I am fairly confident of is the framing: fit-to-fit consistency is not a vague property of a headphone’s comfort, it is a measurable consequence of its acoustic source impedance, and it should be on the spec sheet.