ASR Headphone Testing and BK 5128 Hats Measurement System

[Robbo99999]

You've got the thrust of it - Chris's model allows you to get an approximation of the combined direct + indirect sound power at the eardrum of a head, in a room, for a speaker, given that you have the reverberation time data of the room, the directivity of the speaker, and a set of characterization HRTFs for the head. This would be essentially equivalent to a higher-resolution version of what Olive and Welti did in 2013, where they measured an ear in the Harman room with their Revel speaker pair - a fair amount of smoothing was required for a smooth response due to the in-room condition there, but with the anechoic data, we can be much more high-res

This, I'll note, isn't approximate to the Harman target itself, but rather the "flat in-room baseline" which the Harman target was created from. From there, a large body of listeners were allowed to adjust two shelf filters (12dB/oct at 105~ and 2500~hz, I believe) to preference for both the speakers and headphones, yielding these adjustments in the 2013 study
View attachment 77314
Subsequently small adjustments were made in the 2015 and 2017 papers, but in premise I would argue that applying the adjustment shelves to a Struck-derived "in-room HRTF" would be appropriate, yes.

Re: the Realizer, I'd kind of argue it's almost the opposite, although the practice of measurement is similar - the Realizer does away entirely with target curve theory and simple says "for a given pair of ears, measured at the canal entrance, what transfer function is required to make circumstance A (headphone listening with a given pair of headphones) match circumstance B (speaker listening with a given speaker set and room)?" - it rather neatly "cuts the knot" of what something "should" sound like.

Cool, so I was in the ballpark in terms of the main fundamentals of understanding what the Chris Struck approach was about. Thing is though, how do we actually go about doing the Chris Struck approach, how do we combine all the elements together......there'd be some kind of formulaic combination to combine it into a Target Frequency Response.....do you know how to combine all the elements so that we could actually create some Target Frequency Response Curves based on rooms & speakers of our choice?

Not sure there is anything new here but here is the transfer function for free-field:
View attachment 77331

And diffused:

View attachment 77332

Ah, that's good, potential Target Frequency Responses that can be used, we could experiment with sticking some Harman Curve Low Shelf style bass boosts on those. And I saw in one of your later posts that B&K and/or Sean will be sending you the Harman Target Frequency Response Curve for the B&K 5128, so I think that's excellent news as I expect it to be the most relevant in terms of producing on average the best headphone EQ's for the majority of people here. In addition to the 3 Target Frequency Response Curves mentioned here there's also that Chris Struck Approach that Mad_Economist has been talking about, so I think that would be an interesting addition if we/someone knows how to implement that approach to create some Target Frequency Response Curves.

 

[zenki]

@amirm Possible to measure IEMs as well?

 

[Soniclife]

A bit disappointed on matching of the two ears/couplers at higher frequencies. Combined with variations in headphones, this may be a bit of a mess.

Will measuring some phones backwards, (L & R swapped) help show how big the issue is? Only suitable for phones which will sit right backwards.

 

[Fluffy]

Will measuring some phones backwards, (L & R swapped) help show how big the issue is? Only suitable for phones which will sit right backwards.

Actually that an interesting point. most headphones feature angled drivers, and intuitively it makes sense. But does the angling actually make any difference? if you put the headphones backwards, the angling will be "wrong", it's interesting what effect does it have.

 

[Soniclife]

most headphones feature angled drivers

Do they? It seems to be a growing trend but most seem to aim for flat, with the ear cup rotating to sit correctly on the head.

I'm sure any that are angled would measure differently, the sound will hit the fake ear differently.

 

[solderdude]

Actually that an interesting point. most headphones feature angled drivers, and intuitively it makes sense. But does the angling actually make any difference? if you put the headphones backwards, the angling will be "wrong", it's interesting what effect does it have.

Only very few headphones I tested had angled drivers or angled pads. The Angle usually is very small anyway (about 5 to 6 degrees).
When reversing these the angle difference between the measurements will be 10-12 degrees. That will be visible in the plots when a pinna is used.
Comparing angled to not angled isn't easy because when you remove the angle (when pads are angled) by pressing down the rear side of the pad one also changes the driver-ear distance and air volume. It would be difficult to say if it were the angle or other changes.
It would be fun to see an angled headphone measured the wrong way around though on this HATS.

 

[Soniclife]

Only very few headphones I tested had angled drivers or angled pads. The Angle usually is very small anyway (about 5 to 6 degrees).
When reversing these the angle difference between the measurements will be 10-12 degrees. That will be visible in the plots when a pinna is used.
Comparing angled to not angled isn't easy because when you remove the angle (when pads are angled) by pressing down the rear side of the pad one also changes the driver-ear distance and air volume. It would be difficult to say if it were the angle or other changes.
It would be fun to see an angled headphone measured the wrong way around though on this HATS.

Have you experimented with putting angled pads on normally flat headphones?

 

[Thomas savage]

Maybe serious headphone listeners need to standardize their ears. We'll need an industry agreed upon larger than life pinna shape everyone uses. It will fit over your ears, and connect at the opening to the ear canal. That reduces differences. Then a simpler EQ for ear canal differences might be possible as I don't see how to create an industry standard ear canal for end users.

If your serious about headphone audio and a committed audiophile I don't see why surgery would be out of the question.

Ya gota do what it takes .

 

[VMAT4]

If your serious about headphone audio and a committed audiophile I don't see why surgery would be out of the question.

Ya gota do what it takes .

We'll need to get a list of approved plastic surgeons for this procedure of course!

 

[Thomas savage]

We'll need to get a list of approved plastic surgeons for this procedure of course!

I smell advising revenue and a new winter holiday home in Tenerife..

 

[NTK]

If your serious about headphone audio and a committed audiophile I don't see why surgery would be out of the question.

Ya gota do what it takes .

Mathematicians call this the Procrustes problem.

 

[bobbooo]

I fear you may not find very much that is interesting in terms of distortion characteristics - "typically" headphones have very low distortion, but even some relatively highly nonlinear models are quite subjectively acclaimed. There's an exception here when you get into driver compression/"clipping" at the low end, of course, but the correlation is far from robust.

Incidentally @amirm, it does look like Sean has a target response for you if you'd like it:

https://twitter.com/i/web/status/1292218883785363458

As an aside, I recall AP has added coherence measurements to their functionality list - any plans for measuring headphones with music?

Are these coherence measurements you mention related to the non-coherent distortion tests using music developed by Steve Temme in this AES paper, and shown by Temme and Olive to have higher correlation with (although still not reliably predicting) listener preference ratings than THD, IMD, or standard multitone distortion in this study?

 

[solderdude]

Have you experimented with putting angled pads on normally flat headphones?

Yes, and vice versa. The problem is that usually pads differ in more than just the angle. In that case, because one changes more than one variable, it is hard to say where changes come from exactly. Different foam, different covering material, different height, different angle, different inner dimensions etc.

Maybe Amir has a headphone with an angled driver or pads that are angled. I have not seen this measured. Can't measure it myself as I am pinnaeless. Perhaps the trouserless man in the newsflash below took it ?

In other news, mad trouserless man arrested on the streets of Seattle after trying to slice off human ears. "I only wanted good ears though , it's for SCIENCE!" he cackled at police.

Having a pinna on the test rig would be essential for such test.
Of course I can listen to the K702 in normal way and put it on my head the wrong way and listen for differences. I would have to swap L and R as well though.

 

[Jimbob54]

Yes, and vice versa. The problem is that usually pads differ in more than just the angle. In that case, because one changes more than one variable, it is hard to say where changes come from exactly. Different foam, different covering material, different height, different angle, different inner dimensions etc.

Maybe Amir has a headphone with an angled driver or pads that are angled. I have not seen this measured. Can't measure it myself as I am pinnaeless. Perhaps the trouserless man in the newsflash below took it ?

Having a pinna on the test rig would be essential for such test.
Of course I can listen to the K702 in normal way and put it on my head the wrong way and listen for differences. I would have to swap L and R as well though.

He only collects the freshest of ears . Yours have too many miles on the clock.

 

[Mad_Economist]

Cool, so I was in the ballpark in terms of the main fundamentals of understanding what the Chris Struck approach was about. Thing is though, how do we actually go about doing the Chris Struck approach, how do we combine all the elements together......there'd be some kind of formulaic combination to combine it into a Target Frequency Response.....do you know how to combine all the elements so that we could actually create some Target Frequency Response Curves based on rooms & speakers of our choice?

Struck's approach is basically an adaptation of the Hopkins-Striker equation, which allows the calculation of net sound power in a field including both direct sound and reverberation - because the absolute level isn't a concern, his output basically serves as a way to average between two HRTFs/frequency responses based on some known parameters. One is a free field HRTF (0 degrees, 30 degrees, whatever), representing the "direct sound", and the other a diffuse field HRTF, for the reflections. The weight in the output assigned to these two HRTFs/FRs is based on the reverberation time of the room used in the model, and the directivity of the speaker.

So in other words, if Amir has:
1: A free field HRTF (or a group of them for an average, although I'm not entirely sold on that pitch)
2: A diffuse field HRTF (you can synthesize one of these from free field data if needed, btw)
3: the reverberation time characteristics of the Harman room (which he does)
4: the directivity characteristics of the Revel speakers Sean used (should be possible to find, I'd think)
He can use Struck's methodology here to "simulate" the Harman room, in premise. Neat concept!

If your serious about headphone audio and a committed audiophile I don't see why surgery would be out of the question.

Ya gota do what it takes .

Interestingly, my understanding of human HRTF adaptation is that this shouldn't be too much of a subjective change - once your brain rebuilds the "HRTF library" it uses to attribute localization, I'd expect subjective timbre to be relatively constant.

Are these coherence measurements you mention related to the non-coherent distortion tests using music developed by Steve Temme in this AES paper, and shown by Temme and Olive to have higher correlation with (although still not reliably predicting) listener preference ratings than THD, IMD, or standard multitone distortion in this study?

Eyup - I mean, at the end of the day, these are all just ways you can look at the behavior of a system by comparing two inputs. ARTA's been able to do this for years, and it looks like Audio-Precision added it a year or two back, and of course Temme referenced it in a paper for B&K back in the 90s to my memory. Not a terribly high-tech trick, but it gets applause at parties, and there's some evidence to support it being a preferable test methodology for transducers.

Plus, it's a nice thing to have in your bag of tricks for when people say "But you just tested with sine waves, and sine waves aren't music".

 

[amirm]

So the Harman curve for 5128 is not available. My BK contact thought I wanted the GRAS one. And Sean responded that they are still working on it.

 

[Mad_Economist]

So the Harman curve for 5128 is not available. My BK contact thought I wanted the GRAS one. And Sean responded that they are still working on it.

That's unfortunate to hear. And he won't even give you a sneak preview at it? Spoilsport

 

[Blumlein 88]

Interestingly, my understanding of human HRTF adaptation is that this shouldn't be too much of a subjective change - once your brain rebuilds the "HRTF library" it uses to attribute localization, I'd expect subjective timbre to be relatively constant.

Yeah, I've a read a couple variations where volunteers had molds placed into their outer ear to change shape. It altered their sense of direction and most especially height. After a few weeks, the brain had recalibrated itself and their directional acuity was as good as ever. Further once the molds were removed it took only minutes for the brain to "know" it was back to the old ear shape and function properly.

So at least with binaural, if the whole industry standardized on a single pinna shape, maybe if you listen to enough hours of music your brain will fix itself. Or maybe you need continuous disruption for a few weeks for that to happen. Would be an interesting experiment. Can listeners over time learn to hear another HRTF correctly while periodically switching between theirs and another one.

 

[Robbo99999]

Struck's approach is basically an adaptation of the Hopkins-Striker equation, which allows the calculation of net sound power in a field including both direct sound and reverberation - because the absolute level isn't a concern, his output basically serves as a way to average between two HRTFs/frequency responses based on some known parameters. One is a free field HRTF (0 degrees, 30 degrees, whatever), representing the "direct sound", and the other a diffuse field HRTF, for the reflections. The weight in the output assigned to these two HRTFs/FRs is based on the reverberation time of the room used in the model, and the directivity of the speaker.

So in other words, if Amir has:
1: A free field HRTF (or a group of them for an average, although I'm not entirely sold on that pitch)
2: A diffuse field HRTF (you can synthesize one of these from free field data if needed, btw)
3: the reverberation time characteristics of the Harman room (which he does)
4: the directivity characteristics of the Revel speakers Sean used (should be possible to find, I'd think)
He can use Struck's methodology here to "simulate" the Harman room, in premise. Neat concept!



Interestingly, my understanding of human HRTF adaptation is that this shouldn't be too much of a subjective change - once your brain rebuilds the "HRTF library" it uses to attribute localization, I'd expect subjective timbre to be relatively constant.



Eyup - I mean, at the end of the day, these are all just ways you can look at the behavior of a system by comparing two inputs. ARTA's been able to do this for years, and it looks like Audio-Precision added it a year or two back, and of course Temme referenced it in a paper for B&K back in the 90s to my memory. Not a terribly high-tech trick, but it gets applause at parties, and there's some evidence to support it being a preferable test methodology for transducers.

Plus, it's a nice thing to have in your bag of tricks for when people say "But you just tested with sine waves, and sine waves aren't music".

So the Harman curve for 5128 is not available. My BK contact thought I wanted the GRAS one. And Sean responded that they are still working on it.

In that case Amir the Struck Approach described by Mad_Economist here could be a way to simulate the Harman Curve for the B&K 5128.....and with greater resolution too, in terms of less smoothing. What do you reckon?

 

[DDF]

I worked with a HATS daily evaluating headsets from suppliers. Its flexibility and anatomical shape made test results variable for headsets expecting a tight seal to the ear (high acoustic impedance). Response below 400 Hz varied significantly trial to trial. It became important to use a calibrated spring loaded force gauge to set the clamping force in a repeatable way and then there’s the complication of choosing the clamping force target (unique for each DUT and varies depending on assumed head width).
But this was 30 years ago and I don’t know what improvements were made to the soft pinna. Food for thought! Measuring headsets in a meaningful way is a royal pita!