Why you can’t compare measurements from two different rigs
A frequency-response curve is not a property of the headphone alone. It is a property ofthe headphone and the fixture it was measured on. Two rigs will hand you two different curves for the same pair, and neither is wrong. So a curve with no rig label tells you less than it appears to — and curves from two rigs must never be put on one axis, subtracted, averaged, or counted in a single “N headphones measured” number.
The rig is part of the measurement
To measure a headphone you put it on an artificial ear: a fixture with a pinna, an ear canal of some length, a sealed volume behind it, and a calibrated microphone at the end. Every one of those is an acoustic component, and every one of them shapes the sound before it ever reaches the graph. A different pinna material flexes differently under clamp. A different canal length resonates at a different frequency. A different coupler volume loads the driver differently in the bass.
None of that is error. It is the rig doing exactly what it was built to do — it just means the number you read off the graph describes a system, and the headphone is only half of it. This is the same reason a curve never looks like a flat line even when a headphone sounds neutral, which is the starting point fortarget curves.
Where rigs disagree — and where they mostly don’t
The disagreement is not spread evenly across the spectrum, and knowing where it lives is most of the skill.
In the bass, everything depends on the seal between the earpad and the fixture. A slightly different pinna compliance, a slightly different clamp, and low end that was there a moment ago leaks away. This is why the same open-back headphone can look bass-shy on one rig and fine on another.
Above roughly 8–10 kHz it is worse. Ear-canal resonances dominate, and different couplers put them in different places. Two rigs can disagree by many decibels up there on an identical headphone. That top octave is the region we already tell people not to over-read on any single graph, and cross-rig comparison is precisely where it bites.
The midband is the honest ground. Roughly 200 Hz to a couple of kHz is where fixtures agree best and where a comparison is least likely to be an artifact of the bench. If someone is going to quote one number at you, that is the range where it means something.
It goes finer than rig-to-rig, too. On our own bench the two microphone capsules are individually calibrated and genuinely differ — at 20 kHz one reads +0.37 dB and the other +0.75 dB. Load the wrong calibration file for a channel and you get a curve that looks completely plausible and is wrong in exactly the octave nobody checks.
The trap: compensation makes incomparable curves look comparable
Here is the failure that actually happens. Two curves from two rigs, each compensated toward a target, land on top of each other looking like a matched set. They are not. Compensation subtracts a shape; it does not remove the fixture from the physics. What it does remove is your ability to see that the two measurements come from different worlds.
A raw curve labelled with its rig is honest and slightly ugly. A compensated curve with the rig label dropped is tidy and quietly misleading. If you are collecting graphs from around the internet, the label matters more than the smoothing.
What we did about it ourselves
This is easier to recommend than to live by, so here is what it cost us.
We hold an archive of 319 headphones measured at HeadRoom Corporation on a Head Acoustics HMS II.3, roughly 2010–2013. It is a real historical record and it is genuinely ours to publish. We published it for a while. In August 2026 we stopped, because it does not meet the standard we hold our own bench to: one unit per model, a head that no longer exists, no per-curve measurer, and — the one that decided it — no record of how the headphone was seated. A curve whose seating you cannot describe is a curve you cannot defend, however good the fixture was.
The bench is a miniDSP EARS Pro, commissioned 3 August 2026, with per-capsule calibration and a fixed gain setting that is itself part of the calibration. Its corpus starts at six headphones and grows one session at a time. Small, and honestly labelled as small — which we think is worth more than a large number measured to a standard we can’t describe.
The rule that governs both: two fixtures never share an axis. Not summed, not averaged, not added into one headline count. Every record we keep carries a mandatory rig identifier, because a curve without its coordinate system is not a measurement — it is a picture of one. The temptation to merge an old archive and a new bench into one impressive number is exactly the temptation this rule exists to defeat, and the cheapest way to resist it turned out to be not publishing the old one at all.
What you can and can’t do across rigs
Reasonable: reading broad shape and direction — bass-forward or bass-light, roughly where a large peak sits, whether a headphone is warm or bright. Comparing two headphones to each other when both were measured on the same rig. Using a graph from any rig as a rough sanity check on what you are hearing.
Not reasonable: subtracting one rig’s curve from another’s. Averaging curves from different fixtures into a “consensus.” Building an EQ that corrects a curve measured on rig A toward a target built on rig B. Quoting a decibel difference between two graphs from different sources as if it were a fact about the headphones.
That last one is the common one, and it is worth being blunt about: a 3 dB difference between two graphs from two sites may be entirely the benches talking to each other.
This is also why Attune voice-matches only within a single baseline. When the curve you have and the curve you want were measured on the same rig, the fixture appears in both and cancels exactly in the difference — the rig stops mattering the moment you stop crossing it. Cross baselines and that cancellation quietly stops being true, so we don’t.
See how Attune matches a curveThe honest bottom line
Ask which rig before you ask what the curve says. Compare within a rig, trust the midband most, treat the top octave as indicative rather than exact, and be suspicious of any tidy comparison that does not name its fixtures. A measurement without its coordinate system is not more objective than listening — it just looks like it is.
Common questions
Why do two measurement rigs give different curves for the same headphone?
Because the rig is part of the measurement. The artificial ear, the pinna, the coupler volume and the microphone calibration all shape the response before it reaches the graph. Change the fixture and you change the numbers, even though the headphone never changed.
Where do rigs disagree the most?
In the bass, where the seal against the artificial ear decides how much low end survives, and above roughly 8–10 kHz, where the ear-canal resonances of different couplers diverge sharply. The midband is where rigs agree best, which is why it is the honest place to compare.
Does compensating to a target make measurements comparable?
No, and this is the trap. Compensation makes two curves look like they belong on the same axis by flattening them toward a common shape. The underlying fixture differences are still in there — they are now just harder to see.
What can you compare across rigs?
Broad shape and direction: whether a headphone is bass-forward, where a big peak sits, how two headphones rank against each other when both were measured on the same rig. What you cannot do is subtract, average, or EQ one rig’s curve toward another rig’s target.
Which rig does Warren Labs measure on?
A miniDSP EARS Pro, commissioned 3 August 2026, is our bench and the baseline for everything we publish. We also hold a HeadRoom Corporation archive on a Head Acoustics HMS II.3 from around 2010–2013, which we keep as a research reference and no longer publish as measurements — a rule we applied to ourselves before recommending it to anyone.