• 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

Robbo99999

Master Contributor
Forum Donor
Joined
Jan 23, 2020
Messages
6,875
Likes
6,673
Location
UK
As I appended to my last message above to metal571, I think that would probably depend on the time frame between the measurements, and whether they were done with the same pair of headphones.

I'm guessin that Harman probably made the two measurements shown below using the same pair of AKG K701's though. Probably over a fairly short space of time. If that is the case, then different amounts of wear, or the other issues related to revisions or manufacturing that metal571 touched on above would probably not be the issue in this particular case... Probably. :)

FAja2kvUUAAvYV3
It does seem strange that the bass levels are different, as you'd think bass measurements would be least affected by different choice of ear design, but of course it's obvious that the same target curve can't be used between the two measurement rigs (which is what's been done in this slide, but I don't think you're referring to that point specifically). I caught a snippet of your earlier posts where you postulated that seal might be compromised on the less uniform sealing surface of the B&K 5128 vs the flat surface of the GRAS 45CA - seems likely however the swivel mechanism of the cups on the K702/701 are very flexible and can move freely in all dimensions to allow for optimal angular contact with the surface, but it's still easier to imagine a better seal being achieved on the GRAS 45CA. Perhaps size/depth of outer ear makes a difference if they are different in spec - for instance it seems the closer the driver is to the outer ear then the more bass you get, as seen in worn pad vs fresh pad K702 measurements where thinner pads = more response below 1kHz (or less treble depending on how you look at it)......so perhaps if the B&K outer ear protrudes less from the head ("less deep") then therefore the outer ear is further away from the driver and therefore perhaps less bass, just an idea.
 
Last edited:
  • Like
Reactions: ADU

solderdude

Grand Contributor
Joined
Jul 21, 2018
Messages
15,891
Likes
35,912
Location
The Neitherlands
my prediction is:
From 5Hz to 100Hz is seal dependent (flat pads will work fine, shaped pads may need some assistance)
Up to 1kHz they will measure very similar.
From 1kHz to 5kHz there will be differences (up to a few dB) depending on angle and ear-driver distance and maybe even the amount of trapped air.
From 6kHz to 9kHz there will be some correlation. From 9kHz upwards there will be no correlation.

The sad truth is that regardless how much money is spent on creating the most perfect (average ?) fixture, so ear canal, pinna, head shape, skin emulation the resulting measurements will only be accurate to that standard measurement rig.
Sure, if every one would use that rig measurements will be in agreement about the raw measurements looking very similar up to a certain frequency but seal issues, positioning differences and or averaging of multiple seatings can still create substantial differences.
Just like when Joe, Hank or Harriet put the same headphones on their head and they might not hear what was promised by the plots and the EQ based on a, specific condition measurement, will still make all headphones sound different to them and most other folks.

Yes, the huge tonal balance differences will be much smaller but they'll also be that based on other fixtures.

To me any 'smooth' correction curve, not per se a target curve which I differentiate between, can never be accurate when used to arrive at an EQ.
At least not above 6kHz or so.

The way I see it is: don't look at 1 measurement result of 1 type of fixture and base an 'exact' EQ on that.
Look around on the web for several measurements made by several folks, preferably on different rigs.
Look for common traits and create an EQ based on that. Most likely you'll be closer to a 'universal' correct tonal balance.
Above 6kHz it will be difficult and may require looking for other ways of EQ.

Just my not so expert, but kind of hands on practical opinion that is not widely shared.
 
Last edited:

MayaTlab

Addicted to Fun and Learning
Joined
Aug 15, 2020
Messages
924
Likes
1,512
To the question, it seems conceivable to me...but bear in mind that there isn't a difference in ear Z between the 5128 and 45CA - or any IEC60318-7 compliant HATS - in the band you're talking about. The ITU-T P57 4.3 ear's Z differs at low and high frequency extremes, but generally matches the older 3.3 standard in the midrange and low to mid treble.

That's what I was understanding indeed. So if the 5128 and 45CA are more or less a constant in that regard in the band of interest (300-5kHz), likely causes should be found elsewhere. But that's where I'm just butting head-on against my non-existent knowledge of acoustics. I can only try to measure headphones in situ and not over-interpret the results I obtain (something which I most likely often fail to do), so I'm just hopping that more knowledgeable people at some point would be able to provide some form of explanation as to the different differences observed in that band for some (but not all) HPs.

Assuming that you are using roughly the same type and shape of artificial ears on both setups, you would think that the drivers in a larger circumaural/over-ear headphone would probably be seated a little more closely to the opening of the ear canal on the more curved surface of the mannikin, than on a flat plate (because the area around the ear on the mannikin head is a bit more recessed). Which would seem to favor a more bassy response on the mannikin, in my experience. Rather than the other way around.

The more curved/bumpy surface of the mannikin head probably also has more opportunities for the headphone to leak though. And if the pads on the headphone are more compressed, they may not fill in the gaps between the simulated head and headphones quite as well. (I think Dr. Olive mentioned the cheek as one possible area of leakage on the 5128 mannikin head. And the fixture does appear to be curved/sloped fairly forward in that area in the above picture. Though it's a bit hard to tell from this particular POV. How realistic that may be though for a human head, I couldn't really say.)

I'd love to see a bit more investigation done on the effect of the geometry around the pinna indeed, which in the case of test rigs, and perhaps even the KEMAR, don't look to me superbly anatomically correct.
Not just in terms of seal, but also in terms of front volume "volume", earcup angle relative the ears, and pad compression evenness.
 
D

Deleted member 16543

Guest
Hello,

You have something fouled up in the thought process here.

We are going to start with a headphone, I have an AKG 712. Let’s put this on my GRAS 45 CA-9 test fixture and equalize it as close as possible to the Harman Target curve. At this point the equalization is locked with an armed guard posted.

Next we put this headphone with the fixed equalization on my flat earth test fixture with calibrated pressure microphones and sweep it with the now fixed equalized APx555 generator output.

We will now have a new flat earth target curve calibrated to the freshly calibrated flat earth test fixture.

Now we take a second set of headphones and place them on the flat earth test fixture and equalize them to the new flat earth test curve.

My premise is that this equalized second set of headphones will test very close to the Harman target curve on my GRAS 45 CA-9 test fixture.

Wrap your head around that.

Thanks DT

Simple. It's YOUR premise. But not what actually happens.
Other people have already tried. To this day, IF you can find some reliable transfer function among rigs at all, it is only up to a certain frequency. The more the rig geometry differs from an anthropomorphic outer ear replica, the lower that frequency.
 
D

Deleted member 16543

Guest
Only partially related. One thing to consider is that if wave propagation inside the eardrum was directional, then its transduction to the middle ear and onwards should have psychoacoustic effects. But there is no biological mechanism for detecting directional waves in the ear canal. Timing, sure, and magnitude. Even if you're right, the evolved mechanism seems to discard that additional information (which limits the variation from person to person).

That's doesn't necessarily follow.
Directional contributions translate into differences in amplitude and phase at the eardrum, and that's what is been transmitted into the middle ear and perceived as a difference.
 
D

Deleted member 16543

Guest
Frankly, I don't think it would seem that way to you if you'd actually tried to do these sorts of in situ canal probe measurements. And if the canal is to be attributed significance at low frequencies, you need to contend with Middlebrooks and explain why under his method, which minimizes potential errors by measuring at two points simultaneously, there isn't anything of note under 14k...


Pinnae differ, and the ways that pinnae interact with a given headphone are individual to the pinna and the headphone (in the same way that the difference between my 0 azimuth, 0 elevation free field HRTF and yours, and the difference between my 45 degree azimuth, 45 degree elevation free field HRTF and yours is not necessarily the same). If you measure the response of circumaural headphones at blocked canals (both on humans and mannequins), you will still see variation which cannot be compensated with a static function.

Who's talking about canal probe measurements? Of course I would ascribe those deviations on positional error myself, if that was the case.
But Baffless did these tests on a mannequin head, at the 'eardrum' vs closed canal, with speakers, making sure he wasn't placing the capsule on and off when doing the closed canal measurements round.
And he still found the same order of magnitude of variations.

Also, 14k is audible, so that's proof enough for me that an anthropomorphic canal is indeed needed, for maximum accuracy.
But for now I'm just talking about the deviations observed down to 3 kHz.
 

platimn

Member
Joined
Aug 19, 2021
Messages
33
Likes
16
Does nobody here or elsewhere have an Orpheus they could lend to Amir? It's surprising how few people in the audiophile sphere own one considering its relatively low price ($45,000) is nothing to scoff at but still attainable for people who will spend $4000-5000 on different TOTL flavors every few years. We are not talking the realm of a supercar here, anybody with a decent income and not a lot of sense can finance that kind of purchase. Certainly for an apex of the hobby that is worth it?
 
D

Deleted member 16543

Guest
Does nobody here or elsewhere have an Orpheus they could lend to Amir? It's surprising how few people in the audiophile sphere own one considering its relatively low price ($45,000) is nothing to scoff at but still attainable for people who will spend $4000-5000 on different TOTL flavors every few years. We are not talking the realm of a supercar here, anybody with a decent income and not a lot of sense can finance that kind of purchase. Certainly for an apex of the hobby that is worth it?

This reminded me of this joke.

A: "you know, if you hadn't smoked all these years you'd be driving a Ferrari now."
B: "Yeah? Well.. Where's your Ferrari?"
 

pozz

Слава Україні
Forum Donor
Editor
Joined
May 21, 2019
Messages
4,036
Likes
6,827
That's doesn't necessarily follow.
Directional contributions translate into differences in amplitude and phase at the eardrum, and that's what is been transmitted into the middle ear and perceived as a difference.
That difference would be essentially constant per individual, but the results would depend on the source. How and from what direction the wave reaches the ear. So the only way I see the effect mattering is if it can be proven that the wavefront differentially affects the surface of the eardrum, and the middle ear registers that with 3D oscillation. Or, in terms of measurement, if the ear canal produced inconsistent results along its length with a probe measurement, even when the probe's position is fixed. Perhaps that would include signal dependency, if you're implying that the canal's effect is nonlinear.
 

platimn

Member
Joined
Aug 19, 2021
Messages
33
Likes
16
This reminded me of this joke.

A: "you know, if you hadn't smoked all these years you'd be driving a Ferrari now."
B: "Yeah? Well.. Where's your Ferrari?"
Speaking from personal experience I've nickel and dimed myself to the tune of several thousand on audio gear already, and I haven't even seen no less heard the veritable TOTL. It seems very feasible that some people are already nearing 6 figures on headphones and related gear alone (for speakers of course all reason goes out of the window). There are not very many enthusiast circles where the technological best is actually for sale to consumers for a simple cash payment, but the HE-1 seems to be one of those. It just surprises me that no one has tried. Even with supercars there are ways to get your hands on the good stuff even if you have to reassemble it from scrap. Audiophiles seem to suffer a kind of dynamic compression in this department.
 

pozz

Слава Україні
Forum Donor
Editor
Joined
May 21, 2019
Messages
4,036
Likes
6,827
The other thing he mentioned is that they no longer calibrate at one frequency and instead use white noise.
Was this already discussed? I saw Jude's note about white noise in the charts for the DCA Stealth. I think he does it to check for seal by setting a given voltage level and looking at the difference in SPL between reseats. But I don't know him to have explained his reasoning anywhere directly.
 

DualTriode

Addicted to Fun and Learning
Joined
Oct 24, 2019
Messages
893
Likes
593
Simple. It's YOUR premise. But not what actually happens.
Other people have already tried. To this day, IF you can find some reliable transfer function among rigs at all, it is only up to a certain frequency. The more the rig geometry differs from an anthropomorphic outer ear replica, the lower that frequency.

Hello,

It sounds like you are arguing against target curves and equalization in general. Along the lines that the sound will change inside the headphones on your own head every time that you adjust the headphone position on your own dome.

I am not claiming that there is a transfer function between or among test fixtures.

I am saying that each test fixture will have its’ very own target curve.

Thanks DT
 
D

Deleted member 16543

Guest
That difference would be essentially constant per individual, but the results would depend on the source. How and from what direction the wave reaches the ear. So the only way I see the effect mattering is if it can be proven that the wavefront differentially affects the surface of the eardrum, and the middle ear registers that with 3D oscillation. Or, in terms of measurement, if the ear canal produced inconsistent results along its length with a probe measurement, even when the probe's position is fixed. Perhaps that would include signal dependency, if you're implying that the canal's effect is nonlinear.

No, it also depends on the frequency. Same individual, same source location, but different source frequency means different amplitude and phase changes at the eardrum.
And by continuously comparing these two between left and right eardrums is how we're able to triangulate the source position in the 3D space. The middle ear is only capable to transmit a monodimensional signal with time (that I'm aware of).
 
D

Deleted member 16543

Guest
Hello,

It sounds like you are arguing against target curves and equalization in general. Along the lines that the sound will change inside the headphones on your own head every time that you adjust the headphone position on your own dome.

I am not claiming that there is a transfer function between or among test fixtures.

I am saying that each test fixture will have its’ very own target curve.

Thanks DT

That's not what you're arguing, given your thought experiment.
And in any case, the only target curve that has any meaning to us (because all this has to sooner or later boil down to how good or bad a pair of HPs or speakers sound) is the one that is associated to a rig that more closely resembles a real head.
You can make a target curve for a flat plane exactly like you described, but how is that of any use to us if the correction EQ found for your headphones with that method don't apply when the surface is that of your own head?
 

Mad_Economist

Addicted to Fun and Learning
Audio Company
Joined
Nov 29, 2017
Messages
543
Likes
1,618
Who's talking about canal probe measurements? Of course I would ascribe those deviations on positional error myself, if that was the case.
But Baffless did these tests on a mannequin head, at the 'eardrum' vs closed canal, with speakers, making sure he wasn't placing the capsule on and off when doing the closed canal measurements round.
And he still found the same order of magnitude of variations.

Also, 14k is audible, so that's proof enough for me that an anthropomorphic canal is indeed needed, for maximum accuracy.
But for now I'm just talking about the deviations observed down to 3 kHz.
The actually peer reviewed work (Hammershøi & Møller) that found variations in that band used canal probes for the drum reference point measurements.

Baffless' method avoids the variation inherent to positioning a probe blind in the canal, but brings with it the issues of positioning a probe on an unfeeling dummy. Having had to do that as well, let me tell you, it's a pretty significant error vector, for the same reason that Amir's 5128 IEM measurements were fraught with leakage issues: nerve endings are helpful.

It's also compounded, of course, by the fact that Baffless', like Hammershøi & Møller's, method requires separate sweeps for the entrance and drum measurements, which is most probably where error crept in, in my opinion (although that assumes very consistent placement of the occluding probe, which is itself troublesome).

I have to ask, how do you square your POV here with the fact that most of the Hammershøi & Møller data shows no variation in that band? Is there any precedent for a directional HRTF feature that is completely absent on some individuals?
I am not claiming that there is a transfer function between or among test fixtures.

I am saying that each test fixture will have its’ very own target curve.
What is the utility of a target curve for a test fixture that does not translate to eardrum SPL on humans?
 

DualTriode

Addicted to Fun and Learning
Joined
Oct 24, 2019
Messages
893
Likes
593
The actually peer reviewed work (Hammershøi & Møller) that found variations in that band used canal probes for the drum reference point measurements.

Baffless' method avoids the variation inherent to positioning a probe blind in the canal, but brings with it the issues of positioning a probe on an unfeeling dummy. Having had to do that as well, let me tell you, it's a pretty significant error vector, for the same reason that Amir's 5128 IEM measurements were fraught with leakage issues: nerve endings are helpful.

It's also compounded, of course, by the fact that Baffless', like Hammershøi & Møller's, method requires separate sweeps for the entrance and drum measurements, which is most probably where error crept in, in my opinion (although that assumes very consistent placement of the occluding probe, which is itself troublesome).

I have to ask, how do you square your POV here with the fact that most of the Hammershøi & Møller data shows no variation in that band? Is there any precedent for a directional HRTF feature that is completely absent on some individuals?

What is the utility of a target curve for a test fixture that does not translate to eardrum SPL on humans?

Hello,

You have forgotten the definition of target curve?

This will remind you.

A Statistical Model that Predicts Listeners’ Preference Ratings of Around-Ear and On-Ear Headphones Sean E. Olive, Todd Welti, and Omid Khonsaripour

Thanks DT
 

Mad_Economist

Addicted to Fun and Learning
Audio Company
Joined
Nov 29, 2017
Messages
543
Likes
1,618
Hello,

You have forgotten the definition of target curve?

This will remind you.

A Statistical Model that Predicts Listeners’ Preference Ratings of Around-Ear and On-Ear Headphones Sean E. Olive, Todd Welti, and Omid Khonsaripour

Thanks DT
Yes, but that statistical model is predicate on your target curve predicting behavior on human heads. A fixture whose behavior does not parallel a human head - such as a flush mounted microphone in a plate, be it flat, convex, or concave - will not predict preference ratings because it cannot predict eardrum SPL.
 
D

Deleted member 16543

Guest
The actually peer reviewed work (Hammershøi & Møller) that found variations in that band used canal probes for the drum reference point measurements.

Baffless' method avoids the variation inherent to positioning a probe blind in the canal, but brings with it the issues of positioning a probe on an unfeeling dummy. Having had to do that as well, let me tell you, it's a pretty significant error vector, for the same reason that Amir's 5128 IEM measurements were fraught with leakage issues: nerve endings are helpful.

It's also compounded, of course, by the fact that Baffless', like Hammershøi & Møller's, method requires separate sweeps for the entrance and drum measurements, which is most probably where error crept in, in my opinion (although that assumes very consistent placement of the occluding probe, which is itself troublesome).

I have to ask, how do you square your POV here with the fact that most of the Hammershøi & Møller data shows no variation in that band? Is there any precedent for a directional HRTF feature that is completely absent on some individuals?

What is the utility of a target curve for a test fixture that does not translate to eardrum SPL on humans?

I'm not getting the 'issue' with Baffless methodology.
He placed the mannequin with a capsule at the end of the canal and measured speakers at a few azimuth directions.
Then placed a capsule at the canal entrance and repeated the measurements.
The only thing subject to positioning errors where the speakers on the second round of measurements, which he took good care in minimizing.
The capsule was touched only once, to place it at the canal entrance.

As for how I explain these errors appearing mainly in the 3-6 kHz band I figure it goes something like this.
The only way that the canal would really not matter, mathematically speaking, is if it was an infinite acoustical impedance.
Since it's not infinite (otherwise no wave would make it inside and we'd all be deaf), the impedance at the canal entrance has to be something frequency dependent, and it isn't much of a stretch for me to think that it would also depend on direction, since when we consider a surface of equal input impedance for a certain frequency at the canal entrance, the wave hits all or just some of it, and mostly all at once or 'smeared' in time depending on the incoming angle. Very simplified, ray-like model of incoming sound.. But just to explain the principles. In reality the smooth transition from pinna to canal and the complex geometry of it all makes it so that a different phase shift (and amplitude) for the same source is seen from the canal point of view, depending on the incoming direction, but maybe in a not so easily predictable way.
Even small variations in the impedance would cause a few dB mismatch. We're talking 3-5 difference dB vs 15-25 dB of resonance effect of the canal.
The frequency range where these deviations have been measured is not casual, in my opinion. It's the range where the canal resonates.
Inside the canal it then takes longer or shorter distance to align the wavefront longitudinally depending on frequency.
 
Last edited by a moderator:

DualTriode

Addicted to Fun and Learning
Joined
Oct 24, 2019
Messages
893
Likes
593
Yes, but that statistical model is predicate on your target curve predicting behavior on human heads. A fixture whose behavior does not parallel a human head - such as a flush mounted microphone in a plate, be it flat, convex, or concave - will not predict preference ratings because it cannot predict eardrum SPL.

Hello,

I just searched AES Convention Paper 9919 where the Harman target curve and development research is presented. There is not one mention, prediction or measurement of SPL at the human eardrum. Imagine that.

I am not convinced. I am kind of a hands on kind of guy.

I am sure that you remain solid in your opinion.

What evidence do you have?

Thanks DT
 

Mad_Economist

Addicted to Fun and Learning
Audio Company
Joined
Nov 29, 2017
Messages
543
Likes
1,618
I'm not getting the 'issue' with Baffless methodology.
He placed the mannequin with a capsule at the end of the canal and measured speakers at a few azimuth directions.
Then placed a capsule at the canal entrance and repeated the measurements.
The only thing subject to positioning errors where the speakers on the second round of measurements, which he took good care in minimizing.
The capsule was touched only once, to place it at the canal entrance.
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 anything at 3-6khz...). From the paper whose figure includes the 3khz effect in discussion, I point you to the conclusion that the actual authors came to, which parallels my own on the subject:
1633922890117.png

Fairly extreme precision in measurements must be demonstrated under Baffless' method to ensure that we are not just "measuring noise" with his variations, and to my knowledge this has not been done. If you'd like to, that would be a service to science.
Even small variations in the impedance would cause a few dB mismatch. We're talking 3-5 difference dB vs 15-25 dB of resonance effect of the canal.
The frequency range where these deviations have been measured is not casual, in my opinion. It's the range where the canal resonates.
Inside the canal it then takes longer or shorter distance to align the wavefront longitudinally depending on frequency.
This is a relatively significant impact to attribute to a hitherto-undocumented directional impact on acoustic Z. Again, it could be, but your conjecture here far outstrips the evidence available, and the field consensus here is that the canal is directionally invariant.

Hello,

I just searched AES Convention Paper 9919 where the Harman target curve and development research is presented. There is not one mention, prediction or measurement of SPL at the human eardrum. Imagine that.

I am not convinced. I am kind of a hands on kind of guy.

I am sure that you remain solid in your opinion.

What evidence do you have?

Thanks DT
All measurements used in the Harman study are on a 45CA, at its ear simulators. Measurements at that point are called "DRP", standing for Drum Reference Point, because the acoustic parameters of the HATS or ear simulator match the SPL at, you guessed it, a human eardrum. If you need evidence of this, I recommend reading the pertinent IEC and ITU-T standards (60318-7 and P57/P58).

If your contention here is that the selection of test fixture to measure headphones for the Harman research is arbitrary and unrelated to the behavior in real human ears, someone should probably tell poor Todd Welti.
 
Last edited:
Top Bottom