HBK Headphone Measurement Talks from Head-Fi and Sean Olive

[preload]

Ah, looks like the meaning of the winking smiley in my previous post wasn't conveyed. If you believe that me saying "Just because they didn't mention it publicly doesn't necessarily mean they weren't using the program (or something similar) internally" (i.e. 'it's a possibility') is equivalent to me assuming the definite 'How to Listen was used in the study', then by that (il)logic you saying "This strongly suggests that there could be differences between the training that was utilized in the published research prior to 2009 and the training that is employed in How to Listen" is equivalent to you assuming 'The training used prior to 2009 was not even similar to How to Listen'. Of course, both accusations of assumption are faulty, hence the ironic winking smiley. In reality, neither of us are assuming anything, just guessing at possibilities, but not ruling anything out. Which is why I said 'maybe they did, maybe they didn't' at the end of my post, which you've chosen to overlook for some reason.


Once again, don't let the perfect be the enemy of the good. Of course there is a lot else wrong with audio reviewers and their methods (not sure why you just said speaker reviewers). But attempting to address one area (listener discrimination and reliability) will be a stepwise improvement along the continuum from a Zeos/Darko listening session to a Harman listening test. Blind listening is relatively difficult to set up properly (especially for headphones), even proper level-matching is hard. How to Listen is all ready-made and waiting however, and there can only be benefits to improving the reviewer's listening abilities. I would like all the things you mentioned done (same set of test tracks is another no-brainer and should of course be done), but listener training is a good first step on that road to better subjective audio testing.

Great, thanks for rehashing your perspective. Also, good luck with your noble quest to convince people that not having completed level 8 of How to Listen is a sin. And also good luck convincing people that doing so will solve an important problem.

The irony, being that you might personally benefit from a different type of "How to Listen" course.

 

[preload]

Hello,

First, you need to understand that the GRAS 45 CA MOD used by Harman to derive the Harman target curve is different than the GRAS 45 CA-10 used by ASR (and by me). They are two completely different test fixtures.

The Harman 45CA Mod uses a different pinna and a RA0045 coupler or ear-simulator. GRAS says that the RA0045 ear simulator has a useful frequency range up to about 10 kHz.

The newer high frequency RA0401/2 version coupler or ear-simulator is potentially more accurate and GRAS says that it extends the useful frequency range to 20 kHz.

The thing is that there is no direct conversion between the old and new versions of the GRAS ear-simulators. Perhaps below 10kHZ where the frequency response and impedance of the two different couplers are similar, they are both close enough to the Harman target response without recalculating everything all over again.

Thanks DT

This is really great info, thanks DT.
If the curves <10khz track pretty closely between the CA-MOD and the CA-10, as you say, there at least should be some way to describe the relationship. Perhaps as a range of +/- dB uncertainty vs frequency?

The reason I would like to know (and I'm guessing this is of universal interest too) is that Amir is releasing a lot of HP measurements which is great, but the superimposed Harman target curve may not actually be correct because it was derived on a measurement system that yields slightly different curves.

... Now imagine if Amir's charts showed the Harman-equivalent target, corrected for his rig? We could finally make eq adjustments to match the Harman curve with great accuracy.

 

[preload]

Possibly more so than with HMS or B&K fixtures, but not quite either. That was mentioned in Sean Olive's presentation (slide 75). Harman has the actual data, we don't .
Industry standards or not we're far from dealing with degrees of precision and accuracy below threshold of audible differences anyway, so personally I don't see much of an objection to use Harman's target as a reference point (and only as a reference point) on GRAS fixtures without the modified pinna (particularly the ones like the 45CA with the exact same geometry around the pinna and everything but the pinna ceteris paribus), as long as it's understood that like all headphones measurements on ear simulators, a degree of inaccuracy comes with the baggage and individual experiences may differ to a degree once you're actually wearing the headphones.

I agree, I think it's totally appropriate to superimpose the Harman target based on Harman measurements instead of the reviewers test rig. And sure the good but imperfect correlation with blind preferences and the known variation in treble/bass quantity mean are an issue.

But that being said, we also know that even small 1-2db corrections to the eq curve can turn an excellent hp into a world class hp. Small eq tweaks that "correct" to the approximate Harman target on the Stealth have also made a big improvement. And for that last reason I'm hoping theres a way to see a corrected Harman target so that more deliberate adjustments can be made.

 

[preload]

Besides the AKG K371, there are headphones costing less than $100 that come close to the Harman Target like this one.
View attachment 156459

Methinks @amirm could measure a JBL 710bt, subtract the measured curve from the Harman-measured curve shown here, and use it to generate a "corrected Harman target" for the ASR gras45 rig. In the name of science, of course.

...and ideally before Jude figures out he can do something similar with his rig.

 

[solderdude]

Alas... that isn't possible. Not even when you do this with a 100 headphones and average it.
The thought seems logical but it is too complex as all headphones interact differently on 2 different pinnas.

 

[Deleted member 16543]

Alas... that isn't possible. Not even when you do this with a 100 headphones and average it.
The thought seems logical but it is too complex as all headphones interact differently on 2 different pinnas.

How many times this simple concept has been repeated? And yet people keep thinking it's just a matter of adding some extra EQ to go from one rig to another..
I have a lot more respect for somebody that uses a measurement rig lacking in accuracy like a mic on a flat surface but recognizes its limits, like you do, than for somebody that uses a $10K+ rig (or $40K+) but overlooks this simple concept.

 

[preload]

Alas... that isn't possible. Not even when you do this with a 100 headphones and average it.
The thought seems logical but it is too complex as all headphones interact differently on 2 different pinnas.

How many times this simple concept has been repeated? And yet people keep thinking it's just a matter of adding some extra EQ to go from one rig to another..
I have a lot more respect for somebody that uses a measurement rig lacking in accuracy like a mic on a flat surface but recognizes its limits, like you do, than for somebody that uses a $10K+ rig (or $40K+) but overlooks this simple concept.

Of course it wouldn’t be easy or straightforward.
However I’d like to point out that Harman has already developed such a conversion between the B&K rig and their own gras45 rig. This compensation curve was even featured on one of Olive’s slides (censored of course). So please excuse me when I don’t immediately accept the “it cannot be done” response, particularly since it’s well, been done.

Also, I bet there are parts of the frequency spectrum that are less sensitive to the rig/pinna/etc variations. Understanding this relationship could perhaps allow for saying the compensation curve is accurate for Rig A within 2db for 20hz-200Hz, within 1 do for 200hz-3khz, and useless above 3khz (numbers made up). That would be helpful, particularly since precise eq in the highest few octaves may not be that important for headphones already tuned to sound good “up there.”

But if the whole idea of measuring headphone FR is to understand how closely it follows (or doesn’t follow) a target curve for the purposes of correlating with perceived SQ or making eq adjustments towards that target, but the target is based on another test rig, then what’s the point?? It becomes another set of measurements for the purposes of measurement and marginal practical value.

 

[Robbo99999]

Of course it wouldn’t be easy or straightforward.
However I’d like to point out that Harman has already developed such a conversion between the B&K rig and their own gras45 rig. This compensation curve was even featured on one of Olive’s slides (censored of course). So please excuse me when I don’t immediately accept the “it cannot be done” response, particularly since it’s well, been done.

Also, I bet there are parts of the frequency spectrum that are less sensitive to the rig/pinna/etc variations. Understanding this relationship could perhaps allow for saying the compensation curve is accurate for Rig A within 2db for 20hz-200Hz, within 1 do for 200hz-3khz, and useless above 3khz (numbers made up). That would be helpful, particularly since precise eq in the highest few octaves may not be that important for headphones already tuned to sound good “up there.”

But if the whole idea of measuring headphone FR is to understand how closely it follows (or doesn’t follow) a target curve for the purposes of correlating with perceived SQ or making eq adjustments towards that target, but the target is based on another test rig, then what’s the point?? It becomes another set of measurements for the purposes of measurement and marginal practical value.

(Harman already determined that it wasn't an accurate thing to do - your 1st paragraph)

 

[solderdude]

Did mr. Olive state they just put 1 (or a few) headphones on 2 different rigs and came up with a translation curve ?

Or did they determine what compensation was needed for another test fixture based on what they already know ?

The dips and peaks on different test fixtures, or even the same ones but with other pinnae, isn't easy to compare with narrow dips at different frequencies and with different headphones.

I think they simply tried to create a curve by measuring headphones (which have been fully characterized) and created a Harman curve for that fixture (which is fundamentally different in quite a few aspects) and that they may have produced a plot showing average differences between the fixtures.

 

[Deleted member 16543]

Of course it wouldn’t be easy or straightforward.
However I’d like to point out that Harman has already developed such a conversion between the B&K rig and their own gras45 rig. This compensation curve was even featured on one of Olive’s slides (censored of course). So please excuse me when I don’t immediately accept the “it cannot be done” response, particularly since it’s well, been done.

Also, I bet there are parts of the frequency spectrum that are less sensitive to the rig/pinna/etc variations. Understanding this relationship could perhaps allow for saying the compensation curve is accurate for Rig A within 2db for 20hz-200Hz, within 1 do for 200hz-3khz, and useless above 3khz (numbers made up). That would be helpful, particularly since precise eq in the highest few octaves may not be that important for headphones already tuned to sound good “up there.”

But if the whole idea of measuring headphone FR is to understand how closely it follows (or doesn’t follow) a target curve for the purposes of correlating with perceived SQ or making eq adjustments towards that target, but the target is based on another test rig, then what’s the point?? It becomes another set of measurements for the purposes of measurement and marginal practical value.

What Robbo said.
And to answer the last question you asked, the thing is that there is no point, unless you find a specific target for each specific rig.
Which is the point @solderdude and I were making.

 

[Robbo99999]

Did mr. Olive state they just put 1 (or a few) headphones on 2 different rigs and came up with a translation curve ?

Or did they determine what compensation was needed for another test fixture based on what they already know ?

The dips and peaks on different test fixtures, or even the same ones but with other pinnae, isn't easy to compare with narrow dips at different frequencies and with different headphones.

I think they simply tried to create a curve by measuring headphones (which have been fully characterized) and created a Harman curve for that fixture (which is fundamentally different in quite a few aspects) and that they may have produced a plot showing average differences between the fixtures.

I don't know solderdude, I just remember one of the two Harman folks saying on here that it wasn't an accurate process. They measured a few headphones I think on each to work out the modified Harman Target for B&K 5128 - they determined it wasn't particularly accurate/valid to do such a conversion process and that the Harman work would instead have to be redone from the ground up using the B&K 5128.......that's what I remember unless the Chinese Whispers of my mind have embellished (possible!).

EDIT: this is the best post from Sean Olive that describes it, from this thread:

HBK Headphone Measurement Talks from Head-Fi and Sean Olive

Small variations in capsule placement and the head's alignment can produce narrow variations in HRTF because, of course, that is how we hear - it's why Middlebrooks' methodology was interesting, he largely controlled for that with simultaneous multi-point measurements (which, again, did not find...

www.audiosciencereview.com

and this one:

HBK Headphone Measurement Talks from Head-Fi and Sean Olive

Been mentioned before but worth repeating. On iPhone>accessibility>AirPods>audio accessibility settings>Headphone Accommodations and change the sound to how you like, https://support.apple.com/en-gb/HT211218?cid=mc-ols-audio-article_ht211218-ios_ui-06042020

www.audiosciencereview.com

At this last linked post Sean Olive says:
"So until someone has done a similar study of subjective and objective headphone measurements made on the B&K 5128, this Harman Target customized for the 5128 is the best solution we have for interpreting what the measurements mean in terms of how the headphones sound."

So I think he's acknowledging that it's not ideal unless done from the ground-up. Couldn't really find any info where he points to how inaccurate the headphone translation process is for determining Harman Target for B&K 5128.

 

[solderdude]

In short: If it were easy it would already have been done.

 

[Deleted member 16543]

In short: If it were easy it would already have been done.

Finding the target would actually be quite easy. No idea why B&K hasn't done that yet.

 

[DualTriode]

Hello All,

If you are most people your very own pinna and ear canal attached to your head are different than mine and different than the GRAS 45 CA-10 or B&K HATS 5128 test fixtures. Why would you want to introduce the bias and errors of an artificial pinna and coupler if a flat plane test fixture could perform better?

Chances are, even if you purchased a perfect Harman Target matching set of headphones, that you would if you could adjust the headphone equalization to match your own psychological preference and your very own pinna and ear canal acoustic peaks notches and dips.

Read Harman’s own, U. Horbach AES Convention Paper 9274.



U. Horbach, "Characterizing the Frequency Response of Headphones—A New Paradigm," Paper 9274, (2015 May.). doi:

Abstract: Traditional headphone measurements suffer from large variations if carried out on human subjects with probe microphones, and standardized couplers introduce additional biases, as concluded in a recent paper. Beyond that, there is no clear indication in literature about what the actual perceived frequency response of a headphone might be. This paper explores new measurement methods that avoid the human body as much as possible by measuring the headphone directly, in an attempt to overcome these restrictions and gain more accuracy. Design principles are described in the second part. A novel, DSP controlled, high-quality headphone is introduced that offers the ability to auto-calibrate its frequency response to the individual who is wearing it.



Thanks DT

My thoughts on the Harman Target curve:

The target Curve is a good middle of the road place to start (Least Sum of Squares), it is not too far away from the preference of most headphone users. After a couple of days of break in your new Harman Curve compliant headphones will be your new normal

 

[preload]

I don't know solderdude, I just remember one of the two Harman folks saying on here that it wasn't an accurate process. They measured a few headphones I think on each to work out the modified Harman Target for B&K 5128 - they determined it wasn't particularly accurate/valid to do such a conversion process and that the Harman work would instead have to be redone from the ground up using the B&K 5128.......that's what I remember unless the Chinese Whispers of my mind have embellished (possible!).

EDIT: this is the best post from Sean Olive that describes it, from this thread:

HBK Headphone Measurement Talks from Head-Fi and Sean Olive

Small variations in capsule placement and the head's alignment can produce narrow variations in HRTF because, of course, that is how we hear - it's why Middlebrooks' methodology was interesting, he largely controlled for that with simultaneous multi-point measurements (which, again, did not find...

www.audiosciencereview.com

and this one:

HBK Headphone Measurement Talks from Head-Fi and Sean Olive

Been mentioned before but worth repeating. On iPhone>accessibility>AirPods>audio accessibility settings>Headphone Accommodations and change the sound to how you like, https://support.apple.com/en-gb/HT211218?cid=mc-ols-audio-article_ht211218-ios_ui-06042020

www.audiosciencereview.com

At this last linked post Sean Olive says:
"So until someone has done a similar study of subjective and objective headphone measurements made on the B&K 5128, this Harman Target customized for the 5128 is the best solution we have for interpreting what the measurements mean in terms of how the headphones sound."

So I think he's acknowledging that it's not ideal unless done from the ground-up. Couldn't really find any info where he points to how inaccurate the headphone translation process is for determining Harman Target for B&K 5128.

Okay, this clears things up a little for me - thanks Robbo.
Harman did exactly what I might do if I had access to their blind preference data on a set of headphones and the ability to re-measure all of these headphones on a new rig. That's a no-brainer because it eliminates the uncertainty of how old rig vs new rig measurements might differ in a non-linear (or complex) way.
BUT that doesn't necessarily mean that it's not possible or valid to attempt to translate one curve to the other, meanwhile understanding the level of variation or uncertainty introduced in doing so.
Here's the thing - while you may not get a perfect transfer curve (particularly in the highest octaves), any "correction" is certainly better than whatever we have now, which is a supposed "target" curve that is misrepresented on everyone's FR charts (except Harman's) because it is based on a different measurement system. And at the very least, a translated target curve could easily be represented not by a thin line, but as a thick line/bar that illustrates the degree of variation/uncertainty when translating the Harman target (measured on a Harman rig) to how that target curve would look on a different rig.
Anyway, I can tell there's zero interest in exploring this - I mean it's much easier to say "it can't be done otherwise it would have been done before." Sounds like good logic to me.

 

[DualTriode]

Hello All,

I have not worked since early Covid.

Now I am going on vacation, that means getting on the train and going somewhere.

Sitting here on the bench I have pairs of AKG k371, AKG k701, AKG k240 and when I get back I will have a pair of JBL Tune 710BT's waiting.

The plan is when I return I will post test plots from the GRAS 45 CA-9 and Flat Plate with calibrated 1/4 inch GRAS pressure microphone test fixtures.

Someone else can do the curve calculations with the data.

Thanks DT

 

[Deleted member 16543]

Hello All,

If you are most people your very own pinna and ear canal attached to your head are different than mine and different than the GRAS 45 CA-10 or B&K HATS 5128 test fixtures. Why would you want to introduce the bias and errors of an artificial pinna and coupler if a flat plane test fixture could perform better?

Because you want to measure headphones in a condition as close as possible to the one where they will be normally used.
Just like anechoic measurements of speakers don't mean much, so measurements of 'just the headphones' are equally pointless.
You put the same speaker in different rooms and the sound will be different.
You put the same headphones on different rigs and the measurements will be different.
And the differences will not be predictable or consistent room to room or rig to rig, with the latter being much more unpredictable.
The best sound you get with speakers is when you EQ to a target after having measured them in the space they will be used in, at the listening position. The belief that the speaker's direct sound response has to be flat is not supported by any logic or observation, because correcting a speaker that has a flat direct response is equivalent to modifying that response. And to make a speaker sound better in a real room, at the listening position, you do exactly that.
That's not to say that a speaker with flat direct response AND great directivity is not to be likely to sound pretty good, as is, in most rooms. The point being, it's not the only way to get a balanced response. And you can, and likely will have to, correct the response to get the balance just right.

So... If a speaker's 'as is' measurement is not really telling you that much, similarly, the belief that measuring 'just the headphone' is important is baseless. Even more so, because the behaviour of headphone drivers changes a lot depending on the acoustical load (different rig, different load), while that of a speaker's driver is practically the same in every room, from small to huge.
Also, measuring 'just the headphone' can't be done with a rig such as the one proposed by you. You think just because a flat plane is geometrically simpler than a pinna that it won't add its own weight in the resulting measurements? If you want to measure 'just the headphones', measure them in an anechoic chamber.
The question is... why?

 

[DualTriode]

Because you want to measure headphones in a condition as close as possible to the one where they will be normally used.
Just like anechoic measurements of speakers don't mean much, so measurements of 'just the headphones' are equally pointless.
You put the same speaker in different rooms and the sound will be different.
You put the same headphones on different rigs and the measurements will be different.
And the differences will not be predictable or consistent room to room or rig to rig, with the latter being much more unpredictable.
The best sound you get with speakers is when you EQ to a target after having measured them in the space they will be used in, at the listening position. The belief that the speaker's direct sound response has to be flat is not supported by any logic or observation, because correcting a speaker that has a flat direct response is equivalent to modifying that response. And to make a speaker sound better in a real room, at the listening position, you do exactly that.
That's not to say that a speaker with flat direct response AND great directivity is not to be likely to sound pretty good, as is, in most rooms. The point being, it's not the only way to get a balanced response. And you can, and likely will have to, correct the response to get the balance just right.

So... If a speaker's 'as is' measurement is not really telling you that much, similarly, the belief that measuring 'just the headphone' is important is baseless. Even more so, because the behaviour of headphone drivers changes a lot depending on the acoustical load (different rig, different load), while that of a speaker's driver is practically the same in every room, from small to huge.
Also, measuring 'just the headphone' can't be done with a rig such as the one proposed by you. You think just because a flat plane is geometrically simpler than a pinna that it won't add its own weight in the resulting measurements? If you want to measure 'just the headphones', measure them in an anechoic chamber.
The question is... why?

Hello,

Why the long winded response?

You did not read the AES Conference Paper. The answers to your questions are in the paper.

Look up the work of U. Horbach PhD.

Dr. Horbach is the real deal Audio Engineer / Scientist at Harman.

Thanks DT

 

[thewas]

The belief that the speaker's direct sound response has to be flat is not supported by any logic or observation, because correcting a speaker that has a flat direct response is equivalent to modifying that response. And to make a speaker sound better in a real room, at the listening position, you do exactly that.

That is not supported by everyone, many people agree that above transition frequency you shouldn't correct direct sound based on listening position measurements if the loudspeaker has a flat direct sound and smooth directivity.

Some comments from Floyd Toole about room curve targets, room EQ and more

Room curve targets Every so often it is good to review what we know about room curves, targets, etc. Almost 50 years of double-blind listening tests have shown persuasively that listeners like loudspeakers with flat, smooth, anechoic on-axis and listening-window frequency responses. Those with...

www.audiosciencereview.com

 

[Robbo99999]

Okay, this clears things up a little for me - thanks Robbo.
Harman did exactly what I might do if I had access to their blind preference data on a set of headphones and the ability to re-measure all of these headphones on a new rig. That's a no-brainer because it eliminates the uncertainty of how old rig vs new rig measurements might differ in a non-linear (or complex) way.
BUT that doesn't necessarily mean that it's not possible or valid to attempt to translate one curve to the other, meanwhile understanding the level of variation or uncertainty introduced in doing so.
Here's the thing - while you may not get a perfect transfer curve (particularly in the highest octaves), any "correction" is certainly better than whatever we have now, which is a supposed "target" curve that is misrepresented on everyone's FR charts (except Harman's) because it is based on a different measurement system. And at the very least, a translated target curve could easily be represented not by a thin line, but as a thick line/bar that illustrates the degree of variation/uncertainty when translating the Harman target (measured on a Harman rig) to how that target curve would look on a different rig.
Anyway, I can tell there's zero interest in exploring this - I mean it's much easier to say "it can't be done otherwise it would have been done before." Sounds like good logic to me.

One of the reasons why I don't think it's a particularly good idea to extrapolate a Harman Curve Target for a measurment rig by comparing the same headphones measured on the two different rigs is my own personal experience. Jaakopasenen with his AutoEQ site has done exactly this by comparing measurements of the same model of headphones on say Innerfidelity's rig with Oratory's GRAS and come up with his own Harman Curve Targets for each of the different rigs - so he came up with a Harman Curve for the Innerfidelity rig for instance. I've tried an Innerfidelity Harman Curve EQ on my K702 and it differs greatly from the EQ created from K702 measured by Oratory on GRAS. For instance the following pic is how my K702 looks as measured on a GRAS if I apply the EQ created from the Innerfidelity "Harman" Measurement:

And following is how the K702 should look when measured on a GRAS and EQ'd to the genuine Harman Target for that rig:

So the differences between the final EQ'd frequency responses in the above two graphs are showing the innacuracies of creating a target curve by comparing headphone measurements of the same model between two different measurement rigs. The differences between the two graphs reflect some unit to unit variation as it's not same unit being measured, but it also highlights the massive innacuracies (or differences) that can occur if you rely on creating a "Harman Target" by just comparing headphone measurements of the same headphone between two different measurement rigs. So in personal experience, as well as theory, I don't really see it as a valid & accurate way of creating a Target Curve for measurement rig.