• WANTED: Happy members who like to discuss audio and other topics related to our interest. Desire to learn and share knowledge of science required. There are many reviews of audio hardware and expert members to help answer your questions. Click here to have your audio equipment measured for free!

HBK Headphone Measurement Talks from Head-Fi and Sean Olive

DualTriode

Addicted to Fun and Learning
Joined
Oct 24, 2019
Messages
893
Likes
593
Fun, learning and comparing I'm game as it is for sharing let's share the costs as well after all it is only € 80.- when I decide to keep it (unlikely).
Looks to be half on-ear half over ear (very little room for ears)
Just ordered it (seems to have been a blue one as well).
Should be here in a few days.
Maybe next Thursday I have time to measure and evaluate.
When I will keep it I will PM you my PP address.

It is a plan then.
 
D

Deleted member 16543

Guest
Hello All,

So there you have it, a strong opinion producing a strong emotional response without evidence to back it up.

Thanks DT

I produced the evidence, but you just ignored it and called it a 'long winded response'
Anyway... I'm done feeding the trolls for today.
Bye bye for now
 

GaryH

Major Contributor
Joined
May 12, 2021
Messages
1,348
Likes
1,804
This would only work for N90Q headphones.
Different models will have different PRTFs.
Unless I'm mistaken about the logic you're following, I think it wouldn't work for a simple reason : there isn't a constant transfer function between listeners or between listeners and HATS for all headphones. The HPTF is both listener / HATS and headphones dependent.
I'm aware that different headphones have different PRTF/HPTF, but I'm specifically talking about the delta between the response of a headphone on a HATS and the response on-head. Are we certain that this delta will be different for different headphones (given the same listener and HATS)?

BTW, some of us are interested in knowing where exactly did you source these graphs "from Oratory" ?
A PM on reddit. I don't see what that has at all to do with this discussion though.

P.S. I'm pleased to see you've decided to engage in respectable discourse with me again sans ad hominem. Hope you keep that up.
 
D

Deleted member 16543

Guest
I'm aware that different headphones have different PRTF/HPTF, but I'm specifically talking about the delta between the response of a headphone on a HATS and the response on-head. Are we certain that this delta will be different for different headphones (given the same listener and HATS)?

Yes. That's the reason why a universal transfer function rig to rig (or rig to head, in this case) can't be found.
Different differences for different headphones.
If this wasn't the case, one could 'translate' the Harman target to a B&K 5128, flat plane rigs would work throughout the entire audio band, and DualTriode would have a point.
 

MayaTlab

Addicted to Fun and Learning
Joined
Aug 15, 2020
Messages
924
Likes
1,512
I'm aware that different headphones have different PRTF/HPTF, but I'm specifically talking about the delta between the response of a headphone on a HATS and the response on-head. Are we certain that this delta will be different for different headphones (given the same listener and HATS)?

Yes, otherwise two headphones equalised to the same target on a HATS would indeed sound the same on someone’s head, which they obviously don’t, even when excluding sample variation, as some members here have had their own samples measured on said HATS.
It’s easy to measure anyway below 1kHz.
 

DualTriode

Addicted to Fun and Learning
Joined
Oct 24, 2019
Messages
893
Likes
593
Yes. That's the reason why a universal transfer function rig to rig (or rig to head, in this case) can't be found.
Different differences for different headphones.
If this wasn't the case, one could 'translate' the Harman target to a B&K 5128, flat plane rigs would work throughout the entire audio band, and DualTriode would have a point.

Hello All,

Perhaps this will help.

Hanging around waiting for this Covid thing to end I paid for access to the AES web site and AES Convention Papers. There is a lot of good information.

Harman’s 2018 Convention Paper 9919 put a nail in the relationship between listener preference and the Harman target curve for over ear head phones.

The testing was finalized using the Virtual Test Method. The test subjects could remember that the large heavy headphones were the AUDEZE LCD 2’s. After all of the test headphones were tester and the frequency responses were quantified a replicator headphone pair of AKG k712 headphones, was equalized to match the Frequency Response of each of the 5 or 6 other headphones tested. The correlation coefficient was r = 0.86. That is correlation with real world meaningful numbers. The tests were completed with the same subjects and test fixture.

This is evidence that a pair of AKG k712 headphones can be equalized to stand in for 5 or 6 other headphones of other brands and model numbers.

“Together these validation studies provide evidence that the virtual headphone method produce valid and meaningful results.” Taken from AES Convention Paper 9919.

Thanks DT
 

Robbo99999

Master Contributor
Forum Donor
Joined
Jan 23, 2020
Messages
6,874
Likes
6,672
Location
UK
Yes, otherwise two headphones equalised to the same target on a HATS would indeed sound the same on someone’s head, which they obviously don’t, even when excluding sample variation, as some members here have had their own samples measured on said HATS.
It’s easy to measure anyway below 1kHz.
Yep, this is true, even my two units of K702's that have both been measured by Oratory don't sound exactly the same when EQ'd exactly to the same curve from measurements specific for each particular unit. The difference between the two headphones is that the new K702 has visibly thinner pads vs my old K702 that has new "genuine" pads purchased from Thomann - both headphones were measured as such by Oratory. It does kind of put into perspective any HATS in terms of accuracy for a person.

EDIT: I don't think a flat plate solves this difference though, because the individual headphone related transfer function will still exist for each person & each headphone - so that's not a variable that can be solved by a measurement on any piece of kit, ie one that is not occurring on your own head (in your own ear).
 
Last edited:

MayaTlab

Addicted to Fun and Learning
Joined
Aug 15, 2020
Messages
924
Likes
1,512
The testing was finalized using the Virtual Test Method. The test subjects could remember that the large heavy headphones were the AUDEZE LCD 2’s. After all of the test headphones were tester and the frequency responses were quantified a replicator headphone pair of AKG k712 headphones, was equalized to match the Frequency Response of each of the 5 or 6 other headphones tested.

This is evidence that a pair of AKG k712 headphones can be equalized to stand in for 5 or 6 other headphones of other brands and model numbers.

“Together these validation studies provide evidence that the virtual headphone method produce valid and meaningful results.” Taken from AES Convention Paper 9919.

Figures 5 and 7 from this paper : https://www.aes.org/e-lib/browse.cfm?elib=16874
and more indirectly 5 of this one : https://www.aes.org/tmpFiles/elib/20211106/16768.pdf
incite me to have a much more nuanced picture of the validity of the virtualisation method as a way to score another individual pair of HPs than as a way to provide a platform to test preferences for different targets / curves and observe statistical trends in general.

In fact, one of the explicit interest of using replicator headphones, mentioned in section 2.2 of the article you mentioned, precisely is the question of HPTF variability, which is thought to be lowest for large, open headphones (something Harman's tests on leakage effects demonstrates for lower frequencies). It would make very little sense to test, for example, the preferred amount of bass response on a K371 : https://www.rtings.com/headphones/1-5/graph#1671/7913
Another way the 2013 paper on targets went further to mitigate that issue is to test the same targets (as measured on an ear simulator) on two different headphones (while they received different scores and different relative scores - something that I think is quite interesting to observe -, the overall trend remained the same).

The correlation coefficient was r = 0.86. That is correlation with real world meaningful numbers.

My understanding is that this is the correlation between the 2018 listening tests and the predictive model, not between the real vs. virtualised headphones' scores. The former is an a posteriori way to check whether the math of the model works or not, that's all. It's a completely different thing from the correlation established in the 2013 virtualisation paper.
 
D

Deleted member 16543

Guest
I don't think a flat plate solves this difference though, because the individual headphone related transfer function will still exist for each person & each headphone - so that's not a variable that can be solved by a measurement on any piece of kit, ie one that is not occurring on your own head (in your own ear).

Of course a flat plane doesn't solve anything. It actually exacerbates these differences.
Recognizing that the not quite exactly anatomically accurate rig used by Harman is indeed capable to achieve a pretty good mimicking effect to turn a pair of headphones into another (but then again your own experience doesn't really exactly align with this), the only way to make it an even better and more reliable method for even more people is to use a completely anthropomorphic rig (or rather as close as you can get to an average of it).

The flat plane approach goes in the opposite direction. It is based on the false assumption that one can 'take away' parts of the complex geometry surrounding the headphones when in normal use and expect the remaining parts to keep contributing to the sound as the other parts where never removed.

This is not algebra. It's solving volume integrals with complex geometric boundaries. Very different animal..

It would be simple to do this test.
Build the flat plane rig (it's probably one of the easiest to DIY).
Then take a pair of headphones you like.
Measure them on that rig.
Then EQ other headphones so that they measure the same on the flat plane rig.

Do they sound exactly the same? No, even the best average anthropomorphic rig won't be able to provide exact results.

But is the difference in sound wider than if you had used the GRAS rig, let alone the B&K one? Absolutely yes.

Does this invalidate ALL information you can derive from a flat plane type or rig? No.

But can we do better than that? Absolutely yes.
 

DualTriode

Addicted to Fun and Learning
Joined
Oct 24, 2019
Messages
893
Likes
593
Of course a flat plane doesn't solve anything. It actually exacerbates these differences.
Recognizing that the not quite exactly anatomically accurate rig used by Harman is indeed capable to achieve a pretty good mimicking effect to turn a pair of headphones into another (but then again your own experience doesn't really exactly align with this), the only way to make it an even better and more reliable method for even more people is to use a completely anthropomorphic rig (or rather as close as you can get to an average of it).

The flat plane approach goes in the opposite direction. It is based on the false assumption that one can 'take away' parts of the complex geometry surrounding the headphones when in normal use and expect the remaining parts to keep contributing to the sound as the other parts where never removed.

This is not algebra. It's solving volume integrals with complex geometric boundaries. Very different animal..

It would be simple to do this test.
Build the flat plane rig (it's probably one of the easiest to DIY).
Then take a pair of headphones you like.
Measure them on that rig.
Then EQ other headphones so that they measure the same on the flat plane rig.

Do they sound exactly the same? No, even the best average anthropomorphic rig won't be able to provide exact results.

But is the difference in sound wider than if you had used the GRAS rig, let alone the B&K one? Absolutely yes.

Does this invalidate ALL information you can derive from a flat plane type or rig? No.

But can we do better than that? Absolutely yes.

Hello All,

You have started all over again at square one, perhaps with a bit more understanding this time.

Measuring the headphone SPL with a microphone array inside the headphone flat plate style and then auto equalizing the headphone input is what Horback does with the Harman State of The Art BRS 3D measurement system.

If you place a headphone that was equalized on a flat plate test fixture on your head all the complex pinna and ear-canal geometry is still attached to your head all neatly tucked inside the headphone cup doing what it always does. The SPL is calibrated inside the headphone cup. Your pinna and ear-canal geometry do what they always do and the SPL at you DRP is what it would be if the headphone was equalized on an artificial ear test fixture

Remember that we do this for fun.

I do have an AP analyzer and software on my bench along with a GRAS 45 CA-9 test fixture. I also have the makings of a flat plate test fixture with a ¼ inch calibrated pressure microphone. I do not have a test budget to prove anything.

I can do some sample / example tests. I will measure FR, distortion and a couple of other variables of a couple of headphones plus equalize to something approaching the Harman target on the GRAS CA 45-9.

Then with the same headphones and same equalization I will measure FR on the flat plate test fixture.

Now I will have an approximation of a curve for the flat plate fixture. I will take another headphone and equalize it close to the new flat plate curve. Then see if the FR measured on the GRAS 45 CA-9 looks anything like the original FR curves.

Next I want to see you step up and make some and share some measurements.

Thanks

dy/dt out of here for days



I typed this earlier I do not want to waste it.



Hello All,

Running through the center of this is a common target curve.

Will the Target Curve satisfy everyone? No.

Will the target curve look the same for every Test Fixture and headphone combination? No.

The Target Curve will be some Least Sum of Squares type of model somewhere down the middle of the measurements. A good place to start.

Add your own personal preference and pinna / ear-canal equalization and you will have your new favorite set of headphones.

I am on vacation. See you in some days.

Thanks DT
 
D

Deleted member 16543

Guest
I will measure FR, distortion and a couple of other variables of a couple of headphones plus equalize to something approaching the Harman target on the GRAS CA 45-9.

Then with the same headphones and same equalization I will measure FR on the flat plate test fixture.

Now I will have an approximation of a curve for the flat plate fixture.

No, you don't.
If this method doesn't work to go from the GRAS to B&K rigs, which are definitely somewhat similar, what makes you think it will work going from the GRAS rig to a piece of plywood?
Also, assuming that you can find a correct target for ONE SPECIFIC pair if headphones on the flat plane rig (this is possible, at least conceptually speaking), and that you can equalize all other headphones to that target on that same rig (doable too, cancellation frequencies not matching aside), you keep missing the simple point that the response at your eardrum will not vary by the same amount for all headphones, once you put them back on your head. They WILL sound different.
This is due to the vast geometric boundary changes that occur when placing headphones on the flat plane rig vs. your head.
You want these changes to be minimized, and that's why you need a pinna and a ear canal in your measurement rig.
So that when you move from a response at the DRP of the rig to the response at your actual eardrum, the differences will be:

A. Minimized throughout the whole audio band for that specific pair of headphones you just equalized.

AND, at the same time,

B. Minimized throughout a vast number of headphones.

Once you understand this you'll be able to go back and revise your assumptions, I'm sure.
 

Robbo99999

Master Contributor
Forum Donor
Joined
Jan 23, 2020
Messages
6,874
Likes
6,672
Location
UK
Hello All,

You have started all over again at square one, perhaps with a bit more understanding this time.

Measuring the headphone SPL with a microphone array inside the headphone flat plate style and then auto equalizing the headphone input is what Horback does with the Harman State of The Art BRS 3D measurement system.

If you place a headphone that was equalized on a flat plate test fixture on your head all the complex pinna and ear-canal geometry is still attached to your head all neatly tucked inside the headphone cup doing what it always does. The SPL is calibrated inside the headphone cup. Your pinna and ear-canal geometry do what they always do and the SPL at you DRP is what it would be if the headphone was equalized on an artificial ear test fixture

Remember that we do this for fun.

I do have an AP analyzer and software on my bench along with a GRAS 45 CA-9 test fixture. I also have the makings of a flat plate test fixture with a ¼ inch calibrated pressure microphone. I do not have a test budget to prove anything.

I can do some sample / example tests. I will measure FR, distortion and a couple of other variables of a couple of headphones plus equalize to something approaching the Harman target on the GRAS CA 45-9.

Then with the same headphones and same equalization I will measure FR on the flat plate test fixture.

Now I will have an approximation of a curve for the flat plate fixture. I will take another headphone and equalize it close to the new flat plate curve. Then see if the FR measured on the GRAS 45 CA-9 looks anything like the original FR curves.

Next I want to see you step up and make some and share some measurements.

Thanks

dy/dt out of here for days



I typed this earlier I do not want to waste it.



Hello All,

Running through the center of this is a common target curve.

Will the Target Curve satisfy everyone? No.

Will the target curve look the same for every Test Fixture and headphone combination? No.

The Target Curve will be some Least Sum of Squares type of model somewhere down the middle of the measurements. A good place to start.

Add your own personal preference and pinna / ear-canal equalization and you will have your new favorite set of headphones.

I am on vacation. See you in some days.

Thanks DT
Well it'll be interesting to see the results of your experiments, but I think we all know that we're gonna be seeing differences in the results once you measure them with flat plate derived EQ back on your GRAS unit, but it'll be interesting to see just how much variation there is for a variety of different headphones. Enjoy your holiday in the meantime.
 

GaryH

Major Contributor
Joined
May 12, 2021
Messages
1,348
Likes
1,804
Yes. That's the reason why a universal transfer function rig to rig (or rig to head, in this case) can't be found.
Different differences for different headphones.
If this wasn't the case, one could 'translate' the Harman target to a B&K 5128, flat plane rigs would work throughout the entire audio band, and DualTriode would have a point.
Yes, otherwise two headphones equalised to the same target on a HATS would indeed sound the same on someone’s head, which they obviously don’t, even when excluding sample variation, as some members here have had their own samples measured on said HATS.
It’s easy to measure anyway below 1kHz.

Ok so the question is, why? I can think of three possible main reasons:

1) Seal differences between headphones on heads/HATS (mostly affecting bass response)
2) Pinna deformation differences by pads that touch the ear
3) Acoustic impedance differences between headphones (I suspect this is partly dependent on 1, and maybe 2)

1 and and 2 could be controlled for by only using headphones that seal well, with pads that do not touch the ear. If after this it's still found that there are significant 'different differences for different headphones', then it would seem 3 is the source of this, but I'm not sure if this has been definitively determined yet.

(Flat plate rigs are of course a different matter due to their acoustic impedance being far from that of a human ear, which explains why they produce innacurate and uncompensateable results.)
 
Last edited:

MayaTlab

Addicted to Fun and Learning
Joined
Aug 15, 2020
Messages
924
Likes
1,512
Ok so the question is, why? I can think of three possible main reasons:

1) Seal differences between headphones on heads/HATS (mostly affecting bass response)
2) Pinna deformation differences by pads that touch the ear
3) Acoustic impedance differences between headphones (I suspect this is partly dependent on 1 and 2)

1 and and 2 could be controlled for by only using headphones that seal well, with pads that do not touch the ear. If after this it's still found that there are significant 'different differences for different headphones', then it would seem 3 is the source of this, but I'm not sure if this has been definitively determined yet.

I have to leave the "why" to knowledgeable people and can only raise additional questions (perhaps comprised in what 3) is dealing with, IDK).

Above a few kHz I believe that we're leaving the "pressure chamber condition" range and that local features start to become increasingly important, so even the angle / position of the drivers / earcup / whatever start to matter. Perhaps this might (or might not, IDK) explain the pretty big 6kHz or so spike that I'm experiencing with all samples of the Hi-X65 I've measured so far, regardless of the mic type : https://www.audiosciencereview.com/...hi-x65-open-back-announced.23380/#post-900059
I can only speculate here, but this among other reasons is why I'd raise the question of the anatomy around the pinna, which is incorrect on all ear simulators so far unless you have Fernando Alonso's neck muscles (and will vary quite a lot between individuals anyway).
Regardless of the type of headphones (even larger, open ones), I tend to take everything above 5kHz or so with a big pinch of salt, regardless of the ear simulator or the operator. It's only when a common feature between all types of ear simulators and across various operators repeatedly occur (ex : Focal's 6kHz spike) that I reliably experience something of the same ilk to some degree (which is also why I greatly value some form of diversity in ear simulators measurements),

With ANC headphones the feedback mechanism, which locks the response below 800-1kHz no matter what - as long as a sufficient level of seal is obtained - but leaves it to vary above, and perhaps combined with some extraneous sensitivity in some regions of the spectrum because of some intentional choices related to the overall acoustic design, may make them more susceptible to variation in the ear canal gain region compared to other headphones : https://www.audiosciencereview.com/...d-fi-and-sean-olive.27017/page-12#post-935561
One of the reasons I believe that it would be preferable to attempt to estimate where this variation is the most important and leave out a few EQ filters to let individuals vary this region to taste.
Rtings will soon publish their full QC45 review, which exhibit channel imbalance in the exact same range where they are the most susceptible to compression, and which IMO can be entirely explained by the above phenomenon (and not necessarily by a gross manufacturing defect per se, as only a couple of mm of compression can make them wildly swing around 2-4kHz, indicating a design very sensitive in that region).
Note here that pad compression is only used as a proxy to estimate susceptibility to several variables at the same time, as it's impossible to control all but one of them to test each one of them independently of the other, so it's not a very good test to determine the cause of what's happening.

And then you add on top of that other idiosyncrasies related to the electronics, which can also bring inaccuracies, for example either because of volume-dependent EQ (Bose), if the ear simulator measurements were made at a different volume you usually listen at (quite likely in my case), or the behaviour of Apple's Adaptive EQ's algorithm when ANC is turned off (or in the case of the AirPods 3 when they don't even have ANC), which requires a broad signal such as noise or music to effectively work, and which may lead to errors when measured with sweeps. It's possible that other ANC headphones have other forms of idiosyncrasies, Rtings's review of the JBL Tour One ran into this problem and apparently couldn't find a solution in time for publication : https://www.rtings.com/headphones/reviews/jbl/tour-one-wireless
 
Last edited:

DualTriode

Addicted to Fun and Learning
Joined
Oct 24, 2019
Messages
893
Likes
593
Hello,

I am still away from the bench for a few days. I have given this some thought. I have noted some things.

Are headphone measurements 100% valid, use the word accurate if you are inclined? Short answer, no.

From conference paper 9919 we see that there is an r = 0.86 correlation between subject preference scores and GRAS CA 45 MOD measurements. If you do the statistics thing and square the 0.86 r value, the maximum validity we can expect is 0.74. For psychological results this is not bad but there is still ample room for other variables to creep in and confound things. However we still have a decent Target Curve.

For those that are interested, say we repeat all the same procedures with the B&K HATS 5128 and get a r = 0.88 correlation and a maximum validity of 0.77. We now have a new, equally valid or slightly better but different Target Curve for the 5128 HATS.

The two Target Curves are close to being equal in predicting subject preference.

This is where we point out that there is no direct conversion between the target curves.



The homework questions are:



Why is there no direct conversion between the target curves?



Why can two different headphones measure the same yet not sound the same?



Thanks DT
 
OP
amirm

amirm

Founder/Admin
Staff Member
CFO (Chief Fun Officer)
Joined
Feb 13, 2016
Messages
44,368
Likes
234,394
Location
Seattle Area
I tend to take everything above 5kHz or so with a big pinch of salt, regardless of the ear simulator or the operator.
5 kHz is a bit low but once you get to 8 kHz or so, the job is done. The tonality of music is well decided by then and variations in seating, fixtures, etc. is too much to judge anything above that.

All of this is evident as I EQ headphones and hear the massive differences below the above limit. It is critical to get that right.
 
OP
amirm

amirm

Founder/Admin
Staff Member
CFO (Chief Fun Officer)
Joined
Feb 13, 2016
Messages
44,368
Likes
234,394
Location
Seattle Area
Why is there no direct conversion between the target curves?
Because it was not a design target? BK 5128 was developed with no regard on this basis. Every change they thought was merited, was put in the fixture. So no wonder that you can't linearly map one to the other.

Further, there is no verification of results of 5128. With all the prior work thrown out the window, what is there to decide if the final outcome is correct in all regards?

The notion of averaging really screws up these designs as that does not represent any reality. Had they found a person whose ears were close to the target they wanted and matched that, then they could perform proper analysis of the simulator vs the real person. But there is no person matching what they built so all is lost there.
 
OP
amirm

amirm

Founder/Admin
Staff Member
CFO (Chief Fun Officer)
Joined
Feb 13, 2016
Messages
44,368
Likes
234,394
Location
Seattle Area
Why can two different headphones measure the same yet not sound the same?
No two headphones measure the same. Even the same headphone doesn't measure the same if you took it off the fixture and put it back on.
 

DualTriode

Addicted to Fun and Learning
Joined
Oct 24, 2019
Messages
893
Likes
593
No two headphones measure the same. Even the same headphone doesn't measure the same if you took it off the fixture and put it back on.

Hello @amirm ,

I am within 0.90 agreement with what you say here.

No one has average ears, hearing or preference.

My thoughts on the Harman Target Curve:

“The Target Curve will be some Least Sum of Squares type of model somewhere down the middle of the measurements. A good place to start.” *

Add your own personal preference and equalization and you will have your new favorite set of headphones.

The Harman Target Curve is based on the average preference of a group of subjects. Unfortunately using the Harman Target Curve allows the average listener to creep into you assumptions, measurements and equalization.

I am sorry to tell you that two different headphones can measure the same within the precession of the test fixture and test procedure. It is entirely possible for two different headphones to measure the same. We are not reading to 6 ½ digits of precision.

To tell the truth, in some ways I am surprised at the lack of precision in the measurement of headphones.



Thanks DT

*The primary variable for the Harman Target Curve is DRP Frequency Response. Many other possible variables are not accounted for, controlled or measured. To think of just a few:

Headphone cup volume,

parallel reflecting surfaces,

the type of foam pad.

These variables and others may account for a large percentage of the variance between the Harman GRAS 45 CA test fixture measurements and other potential test fixtures. This is why there will not be a direct comparison between the Harman GRAS 45 CA MOD Target Curve and other potential test fixtures.
 

GaryH

Major Contributor
Joined
May 12, 2021
Messages
1,348
Likes
1,804
Hello All,

Perhaps this will help.

Hanging around waiting for this Covid thing to end I paid for access to the AES web site and AES Convention Papers. There is a lot of good information.

Harman’s 2018 Convention Paper 9919 put a nail in the relationship between listener preference and the Harman target curve for over ear head phones.

The testing was finalized using the Virtual Test Method. The test subjects could remember that the large heavy headphones were the AUDEZE LCD 2’s. After all of the test headphones were tester and the frequency responses were quantified a replicator headphone pair of AKG k712 headphones, was equalized to match the Frequency Response of each of the 5 or 6 other headphones tested. The correlation coefficient was r = 0.86. That is correlation with real world meaningful numbers. The tests were completed with the same subjects and test fixture.

This is evidence that a pair of AKG k712 headphones can be equalized to stand in for 5 or 6 other headphones of other brands and model numbers.

“Together these validation studies provide evidence that the virtual headphone method produce valid and meaningful results.” Taken from AES Convention Paper 9919.

Thanks DT
My understanding is that this is the correlation between the 2018 listening tests and the predictive model, not between the real vs. virtualised headphones' scores. The former is an a posteriori way to check whether the math of the model works or not, that's all. It's a completely different thing from the correlation established in the 2013 virtualisation paper.
True, they are completely different things, with the correlation from the predictive model paper being r = 0.86, but the correlation for the virtualization study just happened to be very similar with r = 0.85. It's an important point though, touched on before, that the correlation being this high suggests individual PRTF differences (and how they interact with different headphones) are not actually hugely important when it comes to preference. And this study was using the old artificial pinnae which overrepresented leakage, as Sean says in his blog:
Since this paper was written in 2013, we've improved the accuracy of the virtualization in part by developing a custom pinnae for our GRAS 45 CA that better simulates the leakage effects of headphones measured on real human subjects
So the correlation using the new pinnae would be even higher.

I have to leave the "why" to knowledgeable people and can only raise additional questions (perhaps comprised in what 3) is dealing with, IDK).

Above a few kHz I believe that we're leaving the "pressure chamber condition" range and that local features start to become increasingly important, so even the angle / position of the drivers / earcup / whatever start to matter. Perhaps this might (or might not, IDK) explain the pretty big 6kHz or so spike that I'm experiencing with all samples of the Hi-X65 I've measured so far, regardless of the mic type : https://www.audiosciencereview.com/...hi-x65-open-back-announced.23380/#post-900059
I can only speculate here, but this among other reasons is why I'd raise the question of the anatomy around the pinna, which is incorrect on all ear simulators so far unless you have Fernando Alonso's neck muscles (and will vary quite a lot between individuals anyway).
Regardless of the type of headphones (even larger, open ones), I tend to take everything above 5kHz or so with a big pinch of salt, regardless of the ear simulator or the operator. It's only when a common feature between all types of ear simulators and across various operators repeatedly occur (ex : Focal's 6kHz spike) that I reliably experience something of the same ilk to some degree (which is also why I greatly value some form of diversity in ear simulators measurements),

With ANC headphones the feedback mechanism, which locks the response below 800-1kHz no matter what - as long as a sufficient level of seal is obtained - but leaves it to vary above, and perhaps combined with some extraneous sensitivity in some regions of the spectrum because of some intentional choices related to the overall acoustic design, may make them more susceptible to variation in the ear canal gain region compared to other headphones : https://www.audiosciencereview.com/...d-fi-and-sean-olive.27017/page-12#post-935561
One of the reasons I believe that it would be preferable to attempt to estimate where this variation is the most important and leave out a few EQ filters to let individuals vary this region to taste.
Rtings will soon publish their full QC45 review, which exhibit channel imbalance in the exact same range where they are the most susceptible to compression, and which IMO can be entirely explained by the above phenomenon (and not necessarily by a gross manufacturing defect per se, as only a couple of mm of compression can make them wildly swing around 2-4kHz, indicating a design very sensitive in that region).
Note here that pad compression is only used as a proxy to estimate susceptibility to several variables at the same time, as it's impossible to control all but one of them to test each one of them independently of the other, so it's not a very good test to determine the cause of what's happening.

And then you add on top of that other idiosyncrasies related to the electronics, which can also bring inaccuracies, for example either because of volume-dependent EQ (Bose), if the ear simulator measurements were made at a different volume you usually listen at (quite likely in my case), or the behaviour of Apple's Adaptive EQ's algorithm when ANC is turned off (or in the case of the AirPods 3 when they don't even have ANC), which requires a broad signal such as noise or music to effectively work, and which may lead to errors when measured with sweeps. It's possible that other ANC headphones have other forms of idiosyncrasies, Rtings's review of the JBL Tour One ran into this problem and apparently couldn't find a solution in time for publication : https://www.rtings.com/headphones/reviews/jbl/tour-one-wireless
Agreed on the lack of realistic anthropometric rig anatomy around the pinnae (which I reckon explains the B&K 5128's apparent underepresentation of bass that Sean Olive alludes to at the end of his HBK talk). You'd have hoped they would have thought about that with the amount they charge for these things...ANC headphone idiosyncrasies of course add even more variables to an already complex picture.

Why can two different headphones measure the same yet not sound the same?
I suggest reading this.
 
Last edited:
Top Bottom