Why equalising headphones based on some rig measurements usually doesn't work well and an alternative affordable method

[thewas]

Below video is for many an expected confirmation that equalising to some predefined curves based on some measurement rigs doesn't work well for most people due to the significant variation of individual HRTFs, especially in the lower bass and treble region, that's why also companies with large knowledge and research budget like Apple, Bose and Genelec try to compensate for those with either optical 3D scans of the ear anatomy or mic measurement inside the pad cavity. Less expected though for many (including me) is using slow sweeps to manually adjust for individual peaks after correcting the base line response based for example on such classic headphone measurements. A third interesting result is how much frequency response even dominates qualities which were usually thought be independent when equalising on the individual HRTF, like for example making an HD600 which is famous of having a narrow soundstage sounding as wide as the famous for it HD800(S).

Here is also a more detailed LMM summary of above

• (00:04–01:01) The team visits the HRTF lab at Imperial College London’s Dyson School of Engineering, where they measure Head-Related Transfer Functions (HRTF)—how each person’s body, head, and ears filter incoming sound. They insert microphones at the ear canal entrance, effectively turning their heads into measurement microphones to analyze headphones as they are actually heard.
• (01:01–02:30) Measuring headphones directly on a person’s head produces data that looks different from traditional measurement rigs (like GRAS mannequins), but it better reflects real listening because each listener’s anatomy changes how sound behaves.
• (01:50–04:42) A key experiment tested AutoEQ to the Harman target using a Sennheiser HD800S. When the EQ created from a measurement rig was applied and then measured on a real head, the frequency response deviated significantly (e.g., bass roll-off and messy treble), showing that EQ optimized for rigs does not translate reliably to individuals.
• (04:42–06:41) Listening tests confirmed this: the AutoEQ preset sounded acceptable on its own, but switching to personalized EQ based on the listener’s own head measurements revealed a much more natural sound, highlighting flaws in generic EQ profiles.
• (06:41–07:10) The takeaway: chasing perfect alignment with the Harman target curve or obsessing over measurement graphs is misguided because actual results vary greatly depending on the listener’s head and ear anatomy.
• (07:10–09:26) Another experiment EQ’d a Sennheiser HD600 to mimic an HD800S by matching their measured responses on the listener’s head. The result: the HD600 reproduced nearly the same sound characteristics (including soundstage) once the frequency responses matched.
• (09:26–10:21) This demonstrates that many perceived headphone traits (detail, soundstage, timbre) are largely tied to frequency response, though mechanical factors (clamp force, openness) still affect the overall experience.
• (10:35–11:35) Future reviews will combine traditional measurement rigs with “on-head” measurement data, allowing reviewers to show both standardized measurements and how headphones actually behave on their own heads.
• (12:02–14:00) Experiments with tone generators and diffuse-field EQ suggest that manually identifying peaks with test tones works well for correcting frequency spikes, meaning listeners can refine EQ themselves without specialized measurement equipment.
• (15:12–17:47) The video also critiques widespread reliance on Harman research and measurement targets, noting that differences in measurement rigs and human ears make it overly simplistic; the research improved headphone design but doesn’t fully capture real-world listening variability.

I personally hope more researchers and companies will put R&D in this as it is the second big step after the Harman research and target for getting everyone to enjoy uncoloured sound reproduction without the large budget needed for a large hifi room with the corresponding loudspeakers, subwoofers, equalisation and room treatment. We are definitely living exciting times in audio.

 

[olieb]

• (01:50–04:42) A key experiment tested AutoEQ to the Harman target using a Sennheiser HD800S. When the EQ created from a measurement rig was applied and then measured on a real head, the frequency response deviated significantly (e.g., bass roll-off and messy treble), showing that EQ optimized for rigs does not translate reliably to individuals.
• (04:42–06:41) Listening tests confirmed this: the AutoEQ preset sounded acceptable on its own, but switching to personalized EQ based on the listener’s own head measurements revealed a much more natural sound, highlighting flaws in generic EQ profiles.

I can only subscribe to that.
And it does not come as a surprise when one looks at the variability of HRTFs.

They use a microphone setup that inserts a disc with holes and keeps the ear canal open.
Does the video comment on that in comparison to using a blocked ear canal (that's what I use as it is much simpler to do with some silicone putty)?

 

[solderdude]

There are 2 major effects in HRTF for OE*, AE* headphones.
One is the pinna and the other one is ear canal.
For IEM's it is insertion depth, seal and ear canal.

I think it is widely known by now that standard measurement fixtures are only accurate to the standard they comply to and the target that is used.
Your ears may or may not comply to that standard.
You may have a preference that differs from a target it may also fit (fairly) well.

Measurements are a good way to look for the greatest deviations from a standards and to gauge distortion, resonances, compression.
The final proof is in personal listening.

This method deals with the pinna aspect only for OE (maybe works well for AE as well) not the ear canal nor hearing loss.
We know that ear canals differ in length, shape and diameter.

At least it will help with final tuning ... and then we have preference and recordings.

* OE stands for On-Ear and AE stands for Around-Ear.
(NOT OE = over-ear and AE = against-ear )

 

[thewas]

Does the video comment on that in comparison to using a blocked ear canal (that's what I use as it is much simpler to do with some silicone putty)?

They said they will talk about blocked ear canal in the next video of this series (this one is the second one).

I think it is widely known by now that standard measurement fixtures are only accurate to the standard they comply to and the target that is used.
Your ears may or may not comply to that standard.
You may have a preference that differs from a target it may also fit (fairly) well.

Actually most ears don't comply as its just an average/approximation.
Individual tonal banal preference can be also an issue but to make sure it is or isn't the case first someone must eliminate the individual HRTF issue.

 

[solderdude]

Problem is everyone should have a mic + ADC + software to measure and one will certainly get differing results when the headphone is put on the head differently IF they want to know.
So... mic + inserting thingy + ADC + software must be made available.
Probably not a problem when one can buy expensive headphones.

 

[Chagall]

Looking forward to see how they use tone generator/frequency sweeps for personalizing EQ.

My guess is the process should go:
1. EQ to Harman using rig standardized mesurements (Amir, Oratory, autoEQ)
2. EQ peaks with sweeps / tones
3. final broad EQ adjustments with actual music you know according to preference.

 

[Blockader]

They said they will talk about blocked ear canal in the next video of this series (this one is the second one).


Actually most ears don't comply as its just an average/approximation.
Individual tonal banal preference can be also an issue but to make sure it is or isn't the case first someone must eliminate the individual HRTF issue.

Do we want to eliminate the HRTF problem entirely or do we want to design headphone drivers with low enough acoustic impedance that individual ear canal geometry simply stops affecting(ideally) the perceived frequency response at the eardrum? Eliminating HRTF requires a fully closed-loop DSP system with an in ear microphone, which is fascinating but probably too ambitious as a starting point. Making the driver and the overall system more compliant with the ear canal's acoustic impedance feels like the more grounded first step.

What I want is thorough testing of headphones with different artificial acoustic couplers. What I expect is to see is minimum variance in frequency response.

 

[solderdude]

Note that Amir's EQ is already a combination of Amirs listening combined with what he saw on the plots.

Starting from 'detailed' measurements on a certain standard fixture would make sense between 100Hz and 2kHz but below 100Hz seal may be an issue (closed headphones).
Above 3kHz your head is very likely different from the measurements, above 6kHz even more so.

Using sweeps/tones is hard to do as this is also level dependent (the hearing is in any case)
Perhaps narrow band noise could be better.

When listening to music I would not suggest to EQ but rather tune to taste as recording tonality usually differs per album.
In most cases the 'old fashioned' tone control with bass and treble dials is enough to compensate for tonal differences in recordings (if only at different listening levels).

What I expect is to see is minimum variance in frequency response.

Have you watched the video.... ?

 

[thewas]

Perhaps narrow band noise could be better.

It is also a good approach which David Griesinger recommends:

 

[thewas]

Do we want to eliminate the HRTF problem entirely or do we want to design headphone drivers with low enough acoustic impedance that individual ear canal geometry simply stops affecting(ideally) the perceived frequency response at the eardrum? Eliminating HRTF requires a fully closed-loop DSP system with an in ear microphone, which is fascinating but probably too ambitious as a starting point. Making the driver and the overall system more compliant with the ear canal's acoustic impedance feels like the more grounded first step.

What I want is thorough testing of headphones with different artificial acoustic couplers. What I expect is to see is minimum variance in frequency response.

From what I have seen till now for the lower frequencies low or adaptive acoustic impedance solutions work, above I guess the ear geometry or in ear canal mics are needed still.

 

[Zekbone]

This does not imply a flaw with demanding stock Harman tuning, by the way.

 

[thewas]

This does not imply a flaw with demanding stock Harman tuning, by the way.

Of course not, the video praises even its significance, it just shows though also that such a tuning (like every tuning) won't automatically work for everyone (but it is a good baseline for individual adaptation).

 

[ZolaIII]

Psy how we hear is quite researched and confirm to a qualitative sample (100+ equal loudness studies and counting) and they evolved and are evolving. Harman is average of it on average SPL (mid to uper 70's) and that's good enough guideline for lows. Highs will be both influenced with ear chenel and ear shelf, age and state of hearing. It's not always posible to achieve perfect seal and some headphones are impedance dependant. Some sites developed very good methodology explaining all of that and more (including fit, clamp and so on). In the end it depends how EQ is tailored (driver limit and optimisation for it) to which precision and so on. After all you can do so much more even with Auto-EQ.
There is no final solution, nor there should be one. There are different designs to their application and their limits and there will always be. Some in ideal conditions come close to what we even individually can call close to perfect to FR and some other aspects, some are easier to fit, ware and use then others and so on. All if this makes recommending one hard thing to do and untrained, unbiased trying on and sleeping on it is recommend and you still probably won't get there at once. Then there are mods and aftermarket costume pads/tips and so on.

 

[Resolve]

There are 2 major effects in HRTF for OE*, AE* headphones.
One is the pinna and the other one is ear canal.
For IEM's it is insertion depth, seal and ear canal.

I think it is widely known by now that standard measurement fixtures are only accurate to the standard they comply to and the target that is used.
Your ears may or may not comply to that standard.
You may have a preference that differs from a target it may also fit (fairly) well.

Measurements are a good way to look for the greatest deviations from a standards and to gauge distortion, resonances, compression.
The final proof is in personal listening.

This method deals with the pinna aspect only for OE (maybe works well for AE as well) not the ear canal nor hearing loss.
We know that ear canals differ in length, shape and diameter.

At least it will help with final tuning ... and then we have preference and recordings.

* OE stands for On-Ear and AE stands for Around-Ear.
(NOT OE = over-ear and AE = against-ear )

We got both open canal and blocked canal HRTFs from the lab visit.

The way I think about this at the moment is that blocked canal gives you the pinna effects in isolation, without the ear load, while open canal gives you pinna effects in the presence of the ear load. Moreover, we are actually able to identify canal effects in the data even though the measurement is taken at the ear entrance and not at the DRP. Those effects show up as narrowband dips, often around 8khz (2nd) and 12khz (3rd) - but they will be slightly different for each of us given that canals also vary. What's less clear is their magnitude, because they can be perceptually relevant with tone-gen as well.

The initial canal resonance is being captured by the mic, I expect due to the mic design that places the capsule deeper into the ear canal than just the outer portion of the resin disc that holds it in place.

What's interesting about the blocked canal data is that it allows us to use our specific pinna effects for the IEM condition, meaning we no longer need to use just a generic HRTF like JM-1 based on population average data. We've already done some preliminary testing with this and it's... kind of remarkable - yes the ear canals vary, but if we isolate the responses to pinna effects alone, ignoring length mode resonances for the moment, it explains the ways in which two people hear the same IEM different to one another, and some of the differences are absolutely wild.

We'll post a video about this once we have some more data.

 

[solderdude]

I assume you already did a 'sweep' to determine your ear resonances.
I also assume you did the same with the mic (partially) blocking the ear canal.
Is the 'found' peaking of the ear canal quite similar between with and without the little mic ?
Did you also block the open part of the mic as an experiment vs not blocked (but same mic) ?

 

[thewas]

Some examples are shown in more detail in this new video:

 

[nereis]

listening to frequency sweeps I am always tempted to equalize for age related hearing loss / equal loudness at modest listening volume in the high frequencies above 10khz with a massive high shelf, given how much headroom there is from having an amp with too much gain, but I am afraid this might generate excessive heat in the coils or having the high frequencies run +30dB hotter than ~1khz might do more to damage my hearing, I wonder if there is any consensus to safe practices to equalizing the frequency extremes.

for sub bass eq over exertion of driver cause sharp rise in distortion and this is easily audible so it is easy to avoid, but for high frequencies not so much...

 

[ZolaIII]

Some examples are shown in more detail in this new video:

That's just classic... Auto-EQ dosent care about funny guys who are bit lost and talk gibberish nor it does about target. It on you to fiddle that part and Harman is just average in room uper mid 70 dB SPL of what natural speakers in room should be (- reverb of course).
Auto-EQ has old school (Butterwort Q default) tone controls and tilt function. Tilt is particularly useful for regulating highs. FIR is well FIR and for high tap one you will need low self latency reciver (including drivers for it).

 

[solderdude]

Auto EQ works fine with speakers in a room.

The complaint is about using targets + measurements from any test fixture and expect the tonal balance from a headphone to sound correct.
Not so much an complaint about auto EQ (would be just as valid with all other EQ derived from measurements) but more about headphone measurements and 'exact' EQ based on certain measurements.
Something I have been advocating here for years b.t.w. and also the reason why Amir's EQ suggestions are not 1:1 based on his measurements.

 

[thewas]

That's just classic... Auto-EQ dosent care about funny guys who are bit lost and talk gibberish nor it does about target. It on you to fiddle that part and Harman is just average in room uper mid 70 dB SPL of what natural speakers in room should be (- reverb of course).
Auto-EQ has old school (Butterwort Q default) tone controls and tilt function. Tilt is particularly useful for regulating highs. FIR is well FIR and for high tap one you will need low self latency reciver (including drivers for it).

That is not the main issue though why it usually doesn't work well for headphones as solderdude well replied above, maybe you should actually watch the video or read at least the summary of the first one in the first post of this thread.