AES 2025 Paper: New targets for the B&K 5128 GRAS 45CA-10

[Mad_Economist]

If the pressure division ratio significant compared to pinna effects?

Is a rock large compared to a tree? Depends on the rock and the tree. For acoustically low-Z designs (e.g. integrated/open front volume headphones and sealed front volume headphones with very low Z drivers), my (limited) experience is that PDR is very low. This matches with the results of Schlieper as well as Hammershøi & Møller. As source Z rises, in theory you will see rising effects there, with the most absurd extreme, in-ear designs, obviously showing no "pinna effect" and a drastic effect of canal load.

 

[Sean Olive]

An alternative that might be considered to Schlieper is a "reference volume" calibration similar to what Per and Gunnar used for calibration of their constant displacement probes for ear Z measurement. Memory serving, Schlieper's impedance tube was not very indicative at the very frequencies where human ear Z varies most

Compared to the other two methods he evaluated (Pressure Division Ratio and Headphone Selection Criterion) the acoustic impedance tube was the most consistent especially at low frequencies.

I had a colleague at Harman Jeon_woo Kim who was an acoustic impedance tube expert and we used them to measure the impedance of in-ear headphones. This information was then used to model and predict the pressure at the DRP of a listener using the feedback microphone. This was another paper presented at the AES Headphone Conference by Etienne Rivet et al. (note: the paper is Open Access = Free)

E-library page - AES

AES E-Library ← Back to search

aes2.org

 

[peniku8]

Is a rock large compared to a tree? Depends on the rock and the tree. For acoustically low-Z designs (e.g. integrated/open front volume headphones and sealed front volume headphones with very low Z drivers), my (limited) experience is that PDR is very low. This matches with the results of Schlieper as well as Hammershøi & Møller. As source Z rises, in theory you will see rising effects there, with the most absurd extreme, in-ear designs, obviously showing no "pinna effect" and a drastic effect of canal load.

So the idea of "HATS with my own pinna" should be feasible for low-Z designs, which would also be beneficial regarding leakage tolerance?

 

[Mad_Economist]

Compared to the other two methods he evaluated (Pressure Division Ratio and Headphone Selection Criterion) the acoustic impedance tube was the most consistent especially at low frequencies.

I had a colleague at Harman Jeon_woo Kim who was an acoustic impedance tube expert and we used them to measure the impedance of in-ear headphones. This information was then used to model and predict the pressure at the DRP of a listener using the feedback microphone. This was another paper presented at the AES Headphone Conference by Etienne Rivet et al. (note: the paper is Open Access = Free)

E-library page - AES

AES E-Library ← Back to search

aes2.org

I would definitely not expect PDR to be highly consistent, given the number of sources of "noise", what I'm proposing is a multi-volume calibration using known acoustic loads, which is what Soren did for the 5128 stuff (section IV) and http://Gunnar and Per did for the original 711 research. If I'm reading Etienne correctly, that was the same method there, but unless I misread Schlieper, it would be different from his method, no?

Edit:

So the idea of "HATS with my own pinna" should be feasible for low-Z designs, which would also be beneficial regarding leakage tolerance?

Yes, but equally for a low-Z design you could simply use a blocked-canal probe mic in your own ears

 

[Sean Olive]

I haven't measured the XO, but the "open" DCA planar designs I've measured were still quite high acoustic Z - indeed, the proof is in the pudding: if output drops in a linear fashion¹ at lower frequencies in the presence of leaks, tautologically the Zout is higher than a design where this is less the case.

1: That is, the FR variation is (relatively) constant versus absolute output level - obviously, in an uncoupled situation, you can also just plain hit the linearity limits of a driver

If I do more headphone preference listening tests on targets or whatever I would choose a reference headphone with low acoustic impedance to minimize acoustic impedance interactions between subject ears and headphones. Probably an electrostatic/HD800 or maybe one of the HiFiMan Unveiled which I am told are consistent across fixtures.

I don't think it was a factor in this paper other than the headphones we used do vary in response across fixtures due to their higher acoustic Z. Of course if all of headphones used to derive targets had low acoustic Z we would have gotten different results and conclusions: probably higher similarity ratings for all 7 targets.

But the conclusions would only be valid for low acoustic Z headphones.

In an ideal world, companies would all design headphones with low acoustic Z to minimize variations across listeners -- but then certain benefits/feature would be sacrificed like isolation from noise. So the other option is to account for the higher acoustic Z effects on frequency response and compensate for them.

 

[fcserei]

I think personalization for individual fit/ear canal acoustics is a good idea beyond giving them a headphone tuned to a target (e.g. HarmaN)

Actually this is what I am actually playing with.

I've made several different scans about my head with different scanners and ran the mesh2HRTF simulation on them. I've got a quite consistent result, a bit different from the standard Kemar and similar HRIRs:

L, R, Real and virtual center speakers in anechoic condition and my diffuse field curve:

if I vector average L, R and phantom Center responses and compare to my diffuse field curve, this is the result:

The effect of center comb filtering is clearly visible.

Let's see what's happening in a ( simulated ) room. I've added multichannel ambisonic room simulations to the direct signal. I've created a couple of virtual dummy head measurements in different virtual acoustic environments with vector averaging L, R, C source signals as they are reaching the ear. I used 2 different small rooms, not unlike the Harman lab where supposedly the Harman curve base measurements were taken, a concert hall, an opera house and a film scoring stage.

Here are the results:

What I see at the first glance:
- bigger acoustic environments are lining up nice with my DFEQ curves (if the best seat of the house is 5-6 m from the stage, you get the diffuse curve )
- the HRIR peak is not affected by the acoustic environment
- my HRIR peak quite bit higher than the average ( 4k vs 2.5-3k ), so maybe matching this peak is what makes headphone sound really personal
- small rooms are an acoustic mess, not only in the bass, but also up to 1-2 kHz (green and purple )
- most acoustic environment produces 6-10db bass boost below 200 Hz, no wonder for the preference for this in the Harman curve
- bass boost is dependent on the material, Opera houses do not have that much ( tried to simulate more than one ), concert halls love it ( for the impact? )


I think the focus should not be on one single curve, although I understand the good sound for the masses point of view. If we are looking for better headphone experience, the personal HRIR and ambience simulation is mandatory in my opinion.

 

[Mad_Economist]

If I do more headphone preference listening tests on targets or whatever I would choose a reference headphone with low acoustic impedance to minimize acoustic impedance interactions between subject ears and headphones. Probably an electrostatic/HD800 or maybe one of the HiFiMan Unveiled which I am told are consistent across fixtures.

I don't think it was a factor in this paper other than the headphones we used do vary in response across fixtures due to their higher acoustic Z. Of course if all of headphones used to derive targets had low acoustic Z we would have gotten different results and conclusions: probably higher similarity ratings for all 7 targets.

Yep, I'm not subtweeting you, just making sure people are aware that "open back" ≠ "low Z". I've seen that take enough times to knee jerk a bit.

But the conclusions would only be valid for low acoustic Z headphones.

In an ideal world, companies would all design headphones with low acoustic Z to minimize variations across listeners -- but then certain benefits/feature would be sacrificed like isolation from noise. So the other option is to account for the higher acoustic Z effects on frequency response and compensate for them.

There's a bit of a nuance there, since feedback-based ANC lowers acoustic Z (for exactly the same reason that negative feedback lowers amplifier output Z). IMO it's viable to make headphones that are acoustically low(er than almost all current designs in) impedance without making huge tradeoffs, given that equalization is cheap and trivially easy these days, but that is, of course, just my opinion on the tradeoffs entailed.

 

[peniku8]

Yes, but equally for a low-Z design you could simply use a blocked-canal probe mic in your own ears

Ah, fair. I forgot that things that become a static influence can just be "removed" from the equation.. and then blocked canal measurements make more sense of course. Still, I suppose the general takeaway is that low-Z=more better

 

[Robbo99999]

There is a new AES paper published by Dr. Sean Olive @Sean Olive and Dan Clark @Dan Clark on potential targets for GRAS 45CA and B&K 5128:
A Preferred Headphone Target Curve Defined for the B&K 5128and GRAS 45CA-10 Test Fixtures

ABSTRACT
The Harman around-ear and on-ear headphone target, defined using a GRAS 45CA test fixture with aRA045 (“711”) ear simulator and custom pinna, is often misused with other fixtures, leading to measurement and interpretation errors due to differing acoustic transfer impedances. We compared several headphones on the original fixture and two newer models, specifying calibration and target curves for each to improve accuracy. The purpose was to define Harman target equivalents for these new fixtures that account for differences in their acoustic impedances and produce similar sound quality. Listening tests assessed how similarly headphones sounded when measured and equalized on all three fixtures using these revised targets.

The paper is a bit difficult to follow. But essentially, they measured a number of headphones on three fixtures: the original modified GRAS used by Harman for its listener preference papers. Then on GRAS 45CA with 5000 pinna (what I use). And B&K 5128. Differences were then incorporated as EQ profile and listeners were asked if they thought the translated targets for 5000 and 5128 matched the response of the original research fixture.

View attachment 483348

The reference itself was in the randomized test, setting the upper bound of similarity at around 90%. Compared to that, GRAS KB5000 single profile for the tested headphone came closest to the reference. B&K 5128 underperformed it some.

Average of 7 headphones reduced similarity to the reference for both fixtures but much more so for B&K 5128 for Noire X headphone. Results are somewhat similar for Noire XO.

In other words, for a specific headphone you can develop a rather accurate target on the new fixtures. But a generalized one based on averages as developed in the original research, does not work that well, and much worse for 5128.

Here are the differences for the 7 headphones under test with each fixture:
View attachment 483350
View attachment 483351

Visually, we can see that the reference research fixture matches new GRAS KB5000 (read and green) quite well up to about 3 kHz to 4 kHz. Above that, KB5000 shows higher treble response than the reference. Fortunately, in my testing, I don't put a ton of value on upper treble and at any rate, adjust filters by ear.

B&K 5120 is a different story. Its response varies based on headphone and has a very different shape from bass up to 10+ kHz. This makes it impossible in my opinion, to create a target for it which translates well from the research reference.

It is kind of fascinating that the DCA Corina produced almost identical response in all three fixtures below 1 kHz.

Please note that all of this is my interpretation of the paper. Actual presentation in next week (Friday or is it Saturday) at the AES conference in Newport Beach.

This is the first such formal study and much needed research to inform us of variations we need to consider in these fixtures. I may develop a new target to overlay on the existing one I use in my reviews with GRAS 45-CA/KB5000.

I'm not sure what method you're gonna use to determine a new target from the information, but just eyeballing those graphs you've put up then if you were gonna choose one of those to represent as the new target then I feel that the Hifiman HE1000 curve looks the most generic & representative to me.

 

[amirm]

I'm not sure what method you're gonna use to determine a new target from the information, but just eyeballing those graphs you've put up then if you were gonna choose one of those to represent as the new target then I feel that the Hifiman HE1000 curve looks the most generic & representative to me.

I am not going to create an all new curve. But rather, consider a boost in 3 to 8 kHz which seems to be common in all of them.

 

[amirm]

It's not impossible to create a target for the 5128, but rather it's "impossible" to create a static direct translation from the harman curve to the 5128, expecting every headphones to sound the same, which is what this paper seems to look into.

I said that the motivation is to create a new target for 5128 that piggybacks on all the Harman research based on another fixture. That is what everyone is trying to do. No one is volunteering to re-run all that research to create a new target for 5128. That it is possible is a guess and academic. Everyone wants a shortcut to those results but that shortcut is not forthcoming. It wasn't when I evaluated the 5128, and it isn't now.

Keep in mind that the box plots you linked are "how similar is the 5128 to the original Harman research fixture", not "which one do you like more" or "which one is the best".

Seems like the plot is lost. As I just explained, that is the motivation, i.e. to find a new target that lets 5128 measurements to correlate with listener preference or what is best as you call. From the paper:

"The purpose of this paper was to define Harman target equivalents for these new test fixtures that account for the acoustic transfer impedance differences and produce similar sound quality."

 

[amirm]

I think the focus should not be on one single curve, although I understand the good sound for the masses point of view.

It is not for the masses but to create a single standard. If all headphones comply with it, and users then have a few dials to taste, we are done. Production would be standardized and we would get out of the hell we are in now where no one knows the tonality of the source content as approved by the artist, label and engineer.

For speakers, we have flat on-axis anechoic response. That may actually not be ideal to Nth degree either. But it has been accepted and good designers strive for it. We need the same for headphones. As far as I am concerned, that is the GRAS 45CA with KB5000. Not a series of curves. And certainly not another one for 5128 that no one can even prove to be correct. It is a cop out to draw big gray borders around some random curve.

I wish there was an effort in AES to create this standard and be done with it. Instead, we have left it to youtubers to invent target curves.

 

[Mad_Economist]

It is not for the masses but to create a single standard. If all headphones comply with it, and users then have a few dials to taste, we are done. Production would be standardized and we would get out of the hell we are in now where no one knows the tonality of the source content as approved by the artist, label and engineer.

For speakers, we have flat on-axis anechoic response. That may actually not be ideal to Nth degree either. But it has been accepted and good designers strive for it. We need the same for headphones. As far as I am concerned, that is the GRAS 45CA with KB5000. Not a series of curves. And certainly not another one for 5128 that no one can even prove to be correct. It is a cop out to draw big gray borders around some random curve.

I wish there was an effort in AES to create this standard and be done with it. Instead, we have left it to youtubers to invent target curves.

Questions of mort de l'auteur aside¹, given how significant the impact of directivity - which is decidedly not standardized among speakers - is upon tone, it seems odd to me to treat speakers as having a single standard which is comparable to what would be implied in trying to nail every headphone to a specific line on the 45CA. To get there, you'd also need to have a specified listening distance (somehow), room reverberation time, and radiation pattern.

I also cannot help but see this as a bit overfitted - we know from the very research being cited that a range of responses with common trends can achieve equivalent preference across the general population, and further that there is at least some clustering of preferences in subgroups such that things within this range may be preferred to each other by a specific group of listeners. It is not a necessity that headphone performance be easily reducible to a single plot which is easily evaluated for "quality".

1: Questions of mort de l'auteur not aside, I'd be intrigued to know what percentage of people would rather have a headphone that makes the music sound "as the mastering engineer heard it" if that response is not the most-preferred for them.

 

[olieb]

I'd be intrigued to know what percentage of people would rather have a headphone that makes the music sound "as the mastering engineer heard it" if that response is not the most-preferred for them.

I doubt that I would hear the same sound as the mastering engineer, even if I would put his/her very headphone on my head. If the fixtures are that different, so will be heads and ears between people.
Therefore we might have a standard (for measurement and comparison) but I do not see how a general target can be had. (Even if not taking into account that there might be different preferences about "good sound".)

 

[fcserei]

I doubt that I would hear the same sound as the mastering engineer, even if I would put his/her very headphone on my head. If the fixtures are that different, so will be different heads and ears.

This is a simulation of what reaches the ear in the same 0pera house with KEMAR vs my personal HRIR ( compared to the Kemar vs personal DFEQ curve)
We don't hear the same way, and we are definitely not isn the same room where the engineer was or the record was made.

Also

If all headphones comply with it, and users then have a few dials to taste,

This is Amir's K371 measurement vs in ear measurements on my head. Quite a lot of dials to turn.

Maybe it would be better to push for using the upcoming technologies ( ANC, in headphone measurement, personalized spatialization...) than trying to find a curve which can not be reproduced in real life anyway.

 

[amirm]

Questions of mort de l'auteur aside¹, given how significant the impact of directivity - which is decidedly not standardized among speakers - is upon tone, it seems odd to me to treat speakers as having a single standard which is comparable to what would be implied in trying to nail every headphone to a specific line on the 45CA. To get there, you'd also need to have a specified listening distance (somehow), room reverberation time, and radiation pattern.

No. You are confusing the environment with what a speaker must do. Best sound reproduction in a room is a two-phased approach: neutral speakers + room optimizations. We do not want speakers to incorporate variations in room response as that is an untenable position.

Further, it is NOT the goal of any such standardization to solve the 100% problem. Such a solution does not exist and cannot exist due to variations in rooms and of course, lack of standardization in music reproduction. What we want to solve is the 70 to 80% of the problem. That is massively better than "anything goes" as has been proven through decades of research by Dr. Toole/Olive et. al. Remember, people used to think every person is different and there is no commonality among them until the research proved them wrong.

Adoption of speaker standards has been stronger, far stronger than what we have in headphones. I think having every youtuber having a finger in the headphone pot, and massive amount of worthless subjective reviews has lent itself to manufacturers not caring a lot about proper standards. But a standard we must have. The rush to buy the next measurement fixture for headphones has made the situation worse.

 

[amirm]

I also cannot help but see this as a bit overfitted - we know from the very research being cited that a range of responses with common trends can achieve equivalent preference across the general population, and further that there is at least some clustering of preferences in subgroups such that things within this range may be preferred to each other by a specific group of listeners. It is not a necessity that headphone performance be easily reducible to a single plot which is easily evaluated for "quality".

I thought I address this already. As has Dr. Toole has said, and I fully agree, listener needs to ultimately be in charge of adapting the tonality to what they want. That does not however, remove the need for standardization and proper response. We need to pick something that has a good chance of being pleasing, or being correct for the lack of a better word, in the cases where such controls don't exist.

I mean look at what people are still doing with bass response:

How much longer do we go with flat bass response??? Sure, a few people like it that way but many more want extra bass.

If we don't rally around one curve, the anything goes like above will continue. Some will be flat, some will be like Harman and some will be overboosted:

The problem therefore is not to go and add other test fixtures in the loop. The problem is resolve the random approach to bass response that we face day in and day out.

 

[amirm]

1: Questions of mort de l'auteur not aside, I'd be intrigued to know what percentage of people would rather have a headphone that makes the music sound "as the mastering engineer heard it" if that response is not the most-preferred for them.

First, you don't know that. Second, we need such a reference to then learn what variation, if any we may like against it. Such data can then be fed to said mastering engineer to have them change their targets. Right now, they run completely open loop, thinking by the very notion of being a "mastering engineer" and having some successful albums, they must have the secret key.

 

[Mad_Economist]

No. You are confusing the environment with what a speaker must do.

No, I am illustrating a point with a metaphor. Do you read what I write before you respond to my messages, or is it free association?

How much longer do we go with flat bass response??? Sure, a few people like it that way but many more want extra bass.

There are acoustic reasons that flat bass response occurs, and "correcting" it acoustically often comes with meaningful tradeoffs. Overall, for passive headphones, it would frequently be far better advised for consumers to buy linear, low acoustic Z designs and adjust bass to preference with EQ. Unless, of course, we are simply optimizing for matching a target.

Further, I had a presentation about this that you're welcome to watch on youtube if you like, but to be brief, the share of people who "like" flat bass response is fairly high, though the well-rated headphones with acoustically flat bass universally had significantly recessed treble. This is no coincidence.

If we don't rally around one curve, the anything goes like above will continue.

Heavens forfend

The problem is resolve the random approach to bass response that we face day in and day out.

The population of listeners equipped with KB50xx pinnae from the factory and perfectly flat-sided heads will be elated by progress on this issue.

First, you don't know that. Second, we need such a reference to then learn what variation, if any we may like against it. Such data can then be fed to said mastering engineer to have them change their targets. Right now, they run completely open loop, thinking by the very notion of being a "mastering engineer" and having some successful albums, they must have the secret key.

"you don't know that" to a hypothetical question - which, I should note, I don't even have a strong prior on, that's legitimately a question that would benefit from empirical inquiry - seems like a datapoint towards the "you do not actually read my posts before you type a response" thesis.

 

[amirm]

"you don't know that" to a hypothetical question - which, I should note, I don't even have a strong prior on, that's legitimately a question that would benefit from empirical inquiry - seems like a datapoint towards the "you do not actually read my posts before you type a response" thesis.

What you post is what is called FUD. "Do we even know if this or that is true?" That is what I addressed. Next time explain your own position before positing it as a way to create doubt.