Resolve's B&K 5128 Headphone Target - you can try the EQ's.....

[Robbo99999]

Yes! Sean had a presentation including this information, and has posted about it on Twitter.

View attachment 282514
For those inclined to look at the difference versus the human average, I've scraped this data using Vituixcad into REW:
View attachment 282518
due to the issues with optically scraping low res information, I advise against "reading the tea leaves" too heavily on the small things and the variation above 15khz - but I think it's amusing to note that commonality in average difference to humans of the KB5010/5011 and 5128!

Note that all measurements were using a blocked meatus microphone, meaning that this reflects differences from the pinna geometry and fit, but not ear Z. We (headphones.com) are planning to do some measurements with non-occluding mics sometime this year to look at how things look when the ear load is added...but there are so many things to do!

I've attached this data as both an mdat (REW format) and .csvs in the attached zipped folder should anyone have an interest in it, with the aforementioned caveat that it's a quick rip of a slide, not the source data.

(One variable being the pool of tested listeners - how large was it, is it representative of world population, or Western population, or Eastern population, etc.) (Age, affect of ear growth, probably not the main relevance). (Before you can be sure about accuracy of said rig to average population.)

 

[abdo123]

Yes! Sean had a presentation including this information, and has posted about it on Twitter.

View attachment 282514
For those inclined to look at the difference versus the human average, I've scraped this data using Vituixcad into REW:
View attachment 282518
due to the issues with optically scraping low res information, I advise against "reading the tea leaves" too heavily on the small things and the variation above 15khz - but I think it's amusing to note that commonality in average difference to humans of the KB5010/5011 and 5128!

Note that all measurements were using a blocked meatus microphone, meaning that this reflects differences from the pinna geometry and fit, but not ear Z. We (headphones.com) are planning to do some measurements with non-occluding mics sometime this year to look at how things look when the ear load is added...but there are so many things to do!

I've attached this data as both an mdat (REW format) and .csvs in the attached zipped folder should anyone have an interest in it, with the aforementioned caveat that it's a quick rip of a slide, not the source data.

This is incredibly appreciated!

looks like Amir's setup is pretty close to Harman's setup like you mentioned, kind of gives me a huge boost in confidence in the measurements conducted here. it might also explains why headphones tuned with algorithms by members here sound a bit on the dull side. I can also now see that the 9 Khz peak is fairly present on every measurement. Science rocks lol.

Also I can understand the concerns with 5128, the differences are for sure more significant with regards to subbass.

Makes the Linus tech tips homebrew target curve look fairly "flat" (no harman sub-bass)

 

[MayaTlab]

Cool, I think I know what you're getting at now, you're saying that in order to do listening testing on headphones that have been equalised to a Target Curve (eg the B&K 5128 in Resolve's project) then it helps massively if you limit it to headphones that don't vary much with spatial reseating on a measurement rig and also don't vary much between people (with both of those things often going hand-in-hand) because this way you can more likely assure that people are experiencing the intended Frequency Response of the Target Curve - it makes the listening tests more accurate. I'd agree with that.

Not quite ! But Perhaps I didn’t make myself understood well enough, sorry for that. That only coupling insensitive headphones should be used for listening tests is a given and a good practice that Harman and others already used.

What I meant is that for these headphones, we know that at least up to 4-5kHz, the on-head, in situ response on the cohort of listeners that participated in listening tests translates pretty well from ear simulators, and that they show good consistency across individuals. Preferences for this or that curve, or adjustments to bass and treble levels were made using these headphones. As a result, given the current lack of listening tests for newer 5128 targets, I believe that it is prudent that preliminary targets produce an error curve close to the one these headphones produce on 711 test rigs compatible with the Harman target, at least up to a few kHz. Otherwise it would logically mean, without any evidence born out of listening tests, that these listeners would have changed preferences in the meantime.

Here’s an illustration of that using three 5128 targets :
- One born out of Harman’s participation to the HBK conference.
- The LMG 0.6 target
- Headphones.com’s 8dB tilt target
I’d like to point out again that all three companies consider these targets preliminary and a work in progress.

And data obtained scrapping the curves from the HBK conference, which has the convenience of using the same samples and Harman’s own modded 45CA, and the inconvenience of being fairly low res.

Here's the 5128 preliminary targets :

This is the error curve of the HD800, HD650 and Utopia, measured on Harman’s modded 45CA, vs the Harman target :

Now this is a comparison of these error curves vs. the three preliminary targets mentioned above :

It should be quite clear that LMG’s target, in spite of being based on DF HRTF, is the closest of the lot overall, and particularly below 1kHz. Which not that surprising, given how it came out to be, as it’s a fairly direct use of Harman’s empirical data in terms of preferential adjustments to the baseline, as unsatisfying as it might be theory-wise. Harman’s own preliminary target misses the mark below 1kHz essentially because they included in the transfer curve a good chunk of headphones that are highly sensitive to coupling issues, which is a fairly obvious methodological issue. It does perform better than the others past 1kHz (which is to be expected given the method used), but this is also where HPTF variation, either inter-individual or inter-HATS, may start to occur, even for large, open over-ears... I'm a bit on the fence here on which approach should be preferred, but as long as targets using other methodologies, such as using DF HRTF as their baseline, don't show an excessive difference in terms of error curve, as a preliminary work it could be OK.

Now if we look at the K550 on the other hand, we see that the LMG target is the worst in terms of matching the 711 Harman target’s error curve :

But that is totally and utterly fine, as the K550 are very sensitive to coupling issues. That the LMG target doesn’t match the 711 error curve is not necessarily indicative of an issue with the 5128 HATS or the LMG target, but in all likelihood indicative of an engineering or design defect on behalf of the K550 (headphones shouldn't be sensitive to coupling issues, it's just bad design and engineering as it makes any attempt at delivering any target futile).

Ulterior listening tests, whether using 711 or 5128 HATS for that matter, could very well prove that an alternative curve, that’s a bit different here and there from the Harman target, is preferred over it, but for now it’s just to me prudent that for these coupling insensitive headphones, the error curve remains similar to what is found on 711 HATS when using the Harman target, and absolutely fine if it isn't the case for headphones sensitive to coupling issues.

 

[MayaTlab]

Did you not read what you quoted from me?

Perhaps I misunderstood what you tried to convey and I'm sorry for that. Maybe you could extend the courtesy though ? When you write this :

yet we have you saying that he must have done that wrong.

This isn't what I wrote. To me being "susceptible to variance in operators' practices" doesn't necessarily mean that one's practice is wrong. That goes for the rest as well.

You have no data to back that.

We do know that it's highly sensitive to coupling issues. I'm a bit bored to go into details but would you say to someone that their own head is faulty just because the in situ FR they experience with the Stealth is markedly different from someone else's or, at the very least, that the difference between them is different from how headphones insensitive to coupling issues translate between them ?

This convo starts to feel like another "where's the evidence that ear pads affect FR ?" moment so perhaps it's best that we end it here. Have a nice WE.

 

[MayaTlab]

My memory from the interview @Resolve and I did with @Sean Olive is that the score delta where you are highly confident that one headphone will be preferred is about 7 points. E.g. 88 > 80 pretty consistently, 84 could be a toss-up with either.

Is that a reference to the RMSE in Harman's articles ?

AES E-Library » A Statistical Model that Predicts Listeners' Preference Ratings of Around-Ear and On-Ear Headphones

A controlled listening test was conducted on 31 different models of around-ear (AE) and on-ear (OE) headphones to determine listeners' sound quality preferences. One-hundred-thirty listeners both trained and untrained rated the headphones based on preference using a virtual headphone method that...

www.aes.org

Generally speaking, I don't understand how we could be highly confident that one singular headphone can be preferred to another if the difference between the predicted score exceeds 7 points, given for a start that a significant number of HPs in that study showed an error between the actual score and the predicted score that exceeded 7.

 

[MayaTlab]

For what it's worth, I do think some experimentation (perhaps making use of some blinded testing? You could convolve some test tracks with IRs and use something like Foobar ABX, although ABX isn't ideal for preference) should be able to yield a most-preferred-for-Robbo response for a given headphone.

I've been using this website for months for ABX tests : https://abx.funkybits.fr/
It includes rating and ranking tests as well, I've dabbled with them for a few weeks already.
What do you think of it ?

 

[MayaTlab]

Has this been documented somewhere?

I'll add to @Mad_Economist's post this one as well :

I know harman research is based on KEMAR, but which pinna？

I am now having a clone pinna for my headphone measurement. And I wanna buy a new pinna, I think if I use the same pinna as in harman research I can get the best comparison results. Yeah, I don't have the head, just have a 43ag like thing. But I think a closer pinna may be better? i don't...

audiosciencereview.com

I wouldn't make a mountain out of a molehill though, particularly given that the most significant differences are going to be at 3kHz+ and remain fairly moderate up to 7kHz at the least.

 

[Robbo99999]

Not quite ! But Perhaps I didn’t make myself understood well enough, sorry for that. That only coupling insensitive headphones should be used for listening tests is a given and a good practice that Harman and others already used.

What I meant is that for these headphones, we know that at least up to 4-5kHz, the on-head, in situ response on the cohort of listeners that participated in listening tests translates pretty well from ear simulators, and that they show good consistency across individuals. Preferences for this or that curve, or adjustments to bass and treble levels were made using these headphones. As a result, given the current lack of listening tests for newer 5128 targets, I believe that it is prudent that preliminary targets produce an error curve close to the one these headphones produce on 711 test rigs compatible with the Harman target, at least up to a few kHz. Otherwise it would logically mean, without any evidence born out of listening tests, that these listeners would have changed preferences in the meantime.

Here’s an illustration of that using three 5128 targets :
- One born out of Harman’s participation to the HBK conference.
- The LMG 0.6 target
- Headphones.com’s 8dB tilt target
I’d like to point out again that all three companies consider these targets preliminary and a work in progress.

And data obtained scrapping the curves from the HBK conference, which has the convenience of using the same samples and Harman’s own modded 45CA, and the inconvenience of being fairly low res.

Here's the 5128 preliminary targets :

View attachment 282533
This is the error curve of the HD800, HD650 and Utopia, measured on Harman’s modded 45CA, vs the Harman target :

View attachment 282535

Now this is a comparison of these error curves vs. the three preliminary targets mentioned above :

View attachment 282536View attachment 282537View attachment 282538

It should be quite clear that LMG’s target, in spite of being based on DF HRTF, is the closest of the lot overall, and particularly below 1kHz. Which not that surprising, given how it came out to be, as it’s a fairly direct use of Harman’s empirical data in terms of preferential adjustments to the baseline, as unsatisfying as it might be theory-wise. Harman’s own preliminary target misses the mark below 1kHz essentially because they included in the transfer curve a good chunk of headphones that are highly sensitive to coupling issues, which is a fairly obvious methodological issue. It does perform better than the others past 1kHz (which is to be expected given the method used), but this is also where HPTF variation, either inter-individual or inter-HATS, may start to occur, even for large, open over-ears... I'm a bit on the fence here on which approach should be preferred, but as long as targets using other methodologies, such as using DF HRTF as their baseline, don't show an excessive difference in terms of error curve, as a preliminary work it could be OK.

Now if we look at the K550 on the other hand, we see that the LMG target is the worst in terms of matching the 711 Harman target’s error curve :
View attachment 282541

But that is totally and utterly fine, as the K550 are very sensitive to coupling issues. That the LMG target doesn’t match the 711 error curve is not necessarily indicative of an issue with the 5128 HATS or the LMG target, but in all likelihood indicative of an engineering or design defect on behalf of the K550 (headphones shouldn't be sensitive to coupling issues, it's just bad design and engineering as it makes any attempt at delivering any target futile).

Ulterior listening tests, whether using 711 or 5128 HATS for that matter, could very well prove that an alternative curve, that’s a bit different here and there from the Harman target, is preferred over it, but for now it’s just to me prudent that for these coupling insensitive headphones, the error curve remains similar to what is found on 711 HATS when using the Harman target, and absolutely fine if it isn't the case for headphones sensitive to coupling issues.

I'm sorry, I really don't understand what you're talking about. I'm not stupid & I know about headphone measurements and a lot of the theory, but I really struggle to understand your posts just very recently (not historically). Maybe you're just not explaining it in a very economical simple way. I might have to revisit your post tomorrow morning, as my mind is a bit more clouded than normal right now. I really don't think headphone measurements are all that complicated when you understand the various facets & explained in an economical manner, but I just don't get your post. (If you like then wait 16hrs before responding to my post to give me a chance to look at it tomorrow morning.)

 

[Mad_Economist]

(This was a post to Amir that you made, but I'm just adding an idea). Well, you guys could also just sit down privately together to have a discussion, you don't have to have the cameras rolling immediately during the whole thing. There's probably quite a bit to be gained from private discussion, and it takes the pressure off it being a Live or recorded event. Well, you could do both, start with a private conversation then agree to a public discussion of which you will know the basic outline.

My main worry there would be the perception that bad faith might arise - a nice part about recorded interactions is that everyone knows that nobody can mischaracterize what transpires, and I know that we've had some pretty rough misunderstandings with @amirm before, which I'd like to avoid repeating.

Is that a reference to the RMSE in Harman's articles ?

View attachment 282544

AES E-Library » A Statistical Model that Predicts Listeners' Preference Ratings of Around-Ear and On-Ear Headphones

A controlled listening test was conducted on 31 different models of around-ear (AE) and on-ear (OE) headphones to determine listeners' sound quality preferences. One-hundred-thirty listeners both trained and untrained rated the headphones based on preference using a virtual headphone method that...

www.aes.org

Generally speaking, I don't understand how we could be highly confident that one singular headphone can be preferred to another if the difference between the predicted score exceeds 7 points, given for a start that a significant number of HPs in that study showed an error between the actual score and the predicted score that exceeded 7.

I suspect it is due to both stdev and rmse converging around 7 points, but if @Sean Olive wants to weigh in it would be much appreciated!

My own interpretation is simply "there would be statistically significant preference for a body of headphones at 88 vs. 80 PPR across a sufficiently sized listener pool" - that may not be Sean's, but I think that's mild enough to consider reasonable!

I've been using this website for months for ABX tests : https://abx.funkybits.fr/
It includes rating and ranking tests as well, I've dabbled with them for a few weeks already.
What do you think of it ?

Neat! I haven't looked at that much, so I'll have to get back to you!

looks like Amir's setup is pretty close to Harman's setup like you mentioned, kind of gives me a huge boost in confidence in the measurements conducted here. it might also explains why headphones tuned with algorithms by members here sound a bit on the dull side. I can also now see that the 9 Khz peak is fairly present on every measurement. Science rocks lol

I would be very cautious about attempting to EQ around features in the 8-10k band - you get con/destructive interference at the canal entrance in this band which will be quite individual to users, headphones, and positioning...

Also I can understand the concerns with 5128, the differences are for sure more significant with regards to subbass.

Although when viewed as absolute difference to the human measurements, in fact we find that they aren't so different!

At low frequencies, this difference will primarily if not solely reflect leakage - i.e. the 5128, with the placements performed by Sean's team, typically had more leakage than human heads, while the 45CA had less. These are both errors, but happily both can be corrected with operator procedure to a pretty meaningful extent. I personally prefer to look at headphone behavior through a lens of "coupled behavior vs. typical on-head behavior vs. normal range of variation due to changing acoustic loads", and am hoping we get into that space in terms of measurement visualization soon...

Makes the Linus tech tips homebrew target curve look fairly "flat" (no harman sub-bass)

One important thing to bear in mind is that these are *not* static features of the rig's response - they are the result of how much leakage occurred in those specific tests. Thus you cannot use the low-frequency difference as any kind of a compensation for differences between systems - it is simply a measure of how much error there was, which is why I flipped it to an absolute difference value above.

At high frequencies, things become more complex there, and that's an area where well controlled experiments are still forthcoming (hopefully), addressing variation in placement "noise".

 

[markanini]

@Mad_Economist @Robbo99999
I suggest adapting a ABCHR test software to blindly compare different EQ settings. Use something arbitrary as the reference, I suggest the output of oratory1990s preset. And use a handful of familiar songs, just so the effort wont be entirely song specific. https://wiki.hydrogenaud.io/index.php?title=ABC/HR

 

[amirm]

I guess we should demand him to redo all the research, then. You know.. to validate his measurements..

I have validated my measurements through listening tests across nearly 100 headphones. Every time I measure a headphone I go through a process of creating an EQ and seeing if the correction relative to target results in better sound. Often I do this testing blind. As long as your expectations are that the target is a guide, then you come out fine. Many people have tried my EQ with similar results, sans minor adjustments.

When you have this volume of testing and public scrutiny for whatever is being cooked up for 5128, then we can talk. Until then, I would stay in the corner and not take shots like this.

 

[amirm]

My main worry there would be the perception that bad faith might arise - a nice part about recorded interactions is that everyone knows that nobody can mischaracterize what transpires, and I know that we've had some pretty rough misunderstandings with @amirm before, which I'd like to avoid repeating.

"Bad faith?" You are inviting me to a public talk with this kind of preconception about me? I have no interest in participating in such a conversation with you then, in public or in private.

You don't need me anyway. You need to see if you can get buy-in from Sean Olive. Only then will I listen to any stories you may have.

 

[Mad_Economist]

"Bad faith?" You are inviting me to a public talk with this kind of preconception about me? I have no interest in participating in such a conversation with you then, in public or in private.

You don't need me anyway. You need to see if you can get buy-in from Sean Olive. Only then will I listen to any stories you may have.

The converse, actually - my perception is that you've been hesitant to discuss this with me/us, and I wanted to ensure that you felt comfortable in the knowledge that we aren't going to do some "Amir exposed" click bait. I haven't seen you engage in that sort of thing, outside of you consistently reading hostility into our communications that isn't intended (e.g. right now).

Edit: to be clear, I at least don't need anyone's individual buy in here, and the drive to have a dialogue here comes from me. I'll happily disagree with Sean, with you, or with Andrew - but I think it's helpful to the community, and ideally just a pleasant time, to sit down for a polite discussion about a topic that's apparently causing some controversy

Edit a deux: to be further clear, I have no concerns about chatting off the record, and if that's preferable, let's do it! I wanted to lead with "here's a way to ensure no risk of skullduggery on my side so that you can come unconcerned", but clearly that hasn't gone well

 

[KiyPhi]

I feel like we're running in circles, so I'll just summarize what I think are the 2 main objections to using the B&K 5128 and my objections to those in turn.

1. The GRAS is a standard and works good enough while the B&K still has to prove it can improve on the results we get with the GRAS.

This is a very non scientific approach to the problem. Standards follow the scientific research, not the other way around. This is obviously a field where standards have been struggling to catch up with the results. Pinna shape, target curve, etc.. all had to be periodically revised to make up for a rig that has some evident limitations, the first one being the 711 coupler is not a good enough representation of a human canal.
Follow good scientific approach and the basic concepts of physics everybody should be familiar with, not standards.

You should re-read what you wrote. Standards follow research. So B&K should not be the standard until the research is done. You keep saying science and physics but I don't think your understanding of those two things is correct in this case. You keep saying physics shows this, science shows that, as if what you are saying is some deductively valid thing but it isn't.

2. Nobody matches the shape of the 5128, so the excercise is pointless.

This is a straw man if I ever saw one. I have stated that the previous standards were very accurate within the range that most people would match and that you would want to show that the B&K was significantly better with research. I also stated that measurements inside your own ear or a replica of your own ear would be valid for you only. Neither one of those is this statement.

The whole point of the excercise is to be able to provide a target that puts us closer to a better starting curve, one that reduces the instances of bad sounding records.

Any curve would technically do this. A curve that is mostly perceptively flat would be the most efficient at this. Harman Target already achieves this, what stops it from working is that industry doesn't adopt the good standard. That's it. If what people listened on is the same tonality as what was mixed on, it would sound better. That is the whole basis for the Harman Target.

According to this reasoning the whole Harman research would also have had to be deemed pointless, even more so because their rig is that much less similar to a human head than the 5128.

You are showing amateur level reasoning here. You beg the question if the improvement of the B&K is significantly better. This is the point that needs to be shown, yet you assume it. If you have this already in your assumptions, discussing with you will go nowhere.

Yet, some decent results were achieved and all that people like Resolve are trying to do is to improve on those results.

And this is fine. But the proper procedure needs to be done. If it turns out to be better, that is good.

This process has to go through the use of a better anatomical match of the rig. Because.. physics. That's enough of a justification.

You have a poor definition of sufficient proof. Explain the physics if you are so sure you are correct.

One doesn't have to prove the 5128 can improve on the results achievable with the GRAS to validate the point I'm making.

Yes it does. If you and they claim it is better, you have to show it is better. You can't just say it. I believe it is Hitchens's razor that states what is claimed without evidence can be dismissed without evidence. If you show no evidence, it can be dismissed without it.

How could one even demand for such a thing?

Because that is how claims work. They need to be supported.

The use one will make of a tool, not the tool itself, determines the validity of the resulting work.
No one single research (use of the tool) can prove the 5128 is better than the GRAS, because there's infinite uses one can make of both tools, the vast majority of which are incorrect.

But we are not talking about infinite uses, we are talking about using them to measure and EQ/tune headphones that aim at a target curve developed on that rig. One will be better than the other for this, and if they are basically the same, might as well stick what what we have because it will allow comparison of what is done later with what has already been done.

With this type of reasoning one can prove that GRAS is better than... GRAS itself! Because there's things that weren't done to the best of accuracy standards with Harman's research (arguably the best use of the GRAS tool attempted.. so far).

I still am not quire sure you know reasoning well. Please show the premises and conclusions using only the reasoning shown that you can prove the GRAS is better than itself. Make sure to use only deductive and valid arguments. I would be amazed to see it done. You can do it traditionally or with predicate calculus.

Obviously, if one can use different research results with the same tool to infer that the tool is better/worse than itself, there has to be something majorly wrong with the concept of blindly attributing qualities of the work to capabilities of the tool.
The B&K 5128 is a better tool. Period.
Time will tell us what we can make out of it, but there is no doubt that its full potential (whether we will be able to fully harness it or not) can bear better fruits than GRAS' full potential (which hasn't been achieved either, to be fair).

Again, needs to be shown before you can accept it. I'm not saying B&K is useless, just that it needs to be researched.

I guess we should demand him to redo all the research, then. You know.. to validate his measurements..

Except the rig Amir uses is compatible with the target and research. There can be some variations in measurements, but as long as you know what those are and don't take the parts that deviate as de-facto accurate, no issues.

 

[Robbo99999]

My main worry there would be the perception that bad faith might arise - a nice part about recorded interactions is that everyone knows that nobody can mischaracterize what transpires, and I know that we've had some pretty rough misunderstandings with @amirm before, which I'd like to avoid repeating.

But there's more potential to feel exposed if you jump straight into a public conversation, I think far better for your initial discussions to be in private so you know where you stand, and then you're in a better position to do a public piece. It does feel like it could get quite personal though, but that's probably more likely to happen if you jump straight into a public video discussion.....I mean I can't imagine a public meltdown (and I'd actually not want to see that) but people don't want to be put into a position where that can be a possibility, talking in private first is better.

 

[solderdude]

I'm not saying B&K is useless, just that it needs to be researched.

That's the key here.

How I see it:

Harman has a 5128, as far as I know (maybe things have changed) there were no plans to validate the 5128 vs their GRAS research so it would appear as that research may not (or may, hopefully in the future) be coming from them.
Remains B&K and I think they are finished and created FF and DF which is what in general these manikins will be used for. I don't think they will be researching audiophile headphone response at all. So I don't expect any 'scientific' research coming from them either.

That leaves some universities and someone in that field would need to think it is important to research the 5128 or...
some guys with experience, lots of audiophile headphones around and access to the different fixtures and a high interest in that particular hobby. And while that is no 'true science' it seems to be the only thing around. Maybe others in the future might also have (different ?) ideas and do their own (hobby) thing with creating their target.
Indeed these will never become actual standards but all that the hobby requires is 'good sound quality' with EQ and good relation between measurements and perceived sound.
The latter is simply impossible but... for some it might and some might even prefer EQ based on that 5128 target and others may prefer one on GRAS (configured in a certain way, using an official standard).

So... just wait and see what becomes of these experiments and what their final conclusion (target) is and perhaps some explanation as to how they arrived at it.
Then, when the EQ based on those sounds good to someone that's good enough as it is just a hobby (headphone music enjoyment) and not designed/targeted at studios as a new 'recording reference target'. Their intention is not to create a new 'standard' or to invalidate research done with other fixtures. It is just to see if it can further improve sound quality and relation to perceived sound. Nothing more that that (AFAICS). One can even participate and converse with those guys and they welcome it (kuddos).

It's just a hobby but can be quite passionate to some. I would like to see what the 'amateurs' come up with in the end and watch the criticism that will follow from all angles and the praise from others claiming the EQ using this target sounds better than with [fill in favorite fixture/person/target] to them.

 

[Robbo99999]

A FP with 5128 pinna+coupler ? or measuring the same headphone on 2 fixtures and match ?

That's not a bad idea. Flat Cheek 5128, and then measuring also on the "round head" 5128 - to expose some models of headphone as having potential seating issues on real heads, and then you've got the flat cheek 5128 to give you the ideal best case measurement. (Resources & extra complication though)

 

[Robbo99999]

Not quite ! But Perhaps I didn’t make myself understood well enough, sorry for that. That only coupling insensitive headphones should be used for listening tests is a given and a good practice that Harman and others already used.

What I meant is that for these headphones, we know that at least up to 4-5kHz, the on-head, in situ response on the cohort of listeners that participated in listening tests translates pretty well from ear simulators, and that they show good consistency across individuals. Preferences for this or that curve, or adjustments to bass and treble levels were made using these headphones. As a result, given the current lack of listening tests for newer 5128 targets, I believe that it is prudent that preliminary targets produce an error curve close to the one these headphones produce on 711 test rigs compatible with the Harman target, at least up to a few kHz. Otherwise it would logically mean, without any evidence born out of listening tests, that these listeners would have changed preferences in the meantime.

Here’s an illustration of that using three 5128 targets :
- One born out of Harman’s participation to the HBK conference.
- The LMG 0.6 target
- Headphones.com’s 8dB tilt target
I’d like to point out again that all three companies consider these targets preliminary and a work in progress.

And data obtained scrapping the curves from the HBK conference, which has the convenience of using the same samples and Harman’s own modded 45CA, and the inconvenience of being fairly low res.

Here's the 5128 preliminary targets :

View attachment 282533
This is the error curve of the HD800, HD650 and Utopia, measured on Harman’s modded 45CA, vs the Harman target :

View attachment 282535

Now this is a comparison of these error curves vs. the three preliminary targets mentioned above :

View attachment 282536View attachment 282537View attachment 282538

It should be quite clear that LMG’s target, in spite of being based on DF HRTF, is the closest of the lot overall, and particularly below 1kHz. Which not that surprising, given how it came out to be, as it’s a fairly direct use of Harman’s empirical data in terms of preferential adjustments to the baseline, as unsatisfying as it might be theory-wise. Harman’s own preliminary target misses the mark below 1kHz essentially because they included in the transfer curve a good chunk of headphones that are highly sensitive to coupling issues, which is a fairly obvious methodological issue. It does perform better than the others past 1kHz (which is to be expected given the method used), but this is also where HPTF variation, either inter-individual or inter-HATS, may start to occur, even for large, open over-ears... I'm a bit on the fence here on which approach should be preferred, but as long as targets using other methodologies, such as using DF HRTF as their baseline, don't show an excessive difference in terms of error curve, as a preliminary work it could be OK.

Now if we look at the K550 on the other hand, we see that the LMG target is the worst in terms of matching the 711 Harman target’s error curve :
View attachment 282541

But that is totally and utterly fine, as the K550 are very sensitive to coupling issues. That the LMG target doesn’t match the 711 error curve is not necessarily indicative of an issue with the 5128 HATS or the LMG target, but in all likelihood indicative of an engineering or design defect on behalf of the K550 (headphones shouldn't be sensitive to coupling issues, it's just bad design and engineering as it makes any attempt at delivering any target futile).

Ulterior listening tests, whether using 711 or 5128 HATS for that matter, could very well prove that an alternative curve, that’s a bit different here and there from the Harman target, is preferred over it, but for now it’s just to me prudent that for these coupling insensitive headphones, the error curve remains similar to what is found on 711 HATS when using the Harman target, and absolutely fine if it isn't the case for headphones sensitive to coupling issues.

Hi, I had another look at this just now this morning, and I'm still finding it confusing. Instead, could you just summarise into a couple of sentences the points you're making, as an outline overview if you will. This way I can probably take what you say and then be able to look at all your detailed graphs with some kind of perspective. If you can, try to avoid jargon such as "error curves". I'm not a laymen, but try to explain it in laymen's terms (or "ASR laymen's terms"), (which should help more than just myself).....I'll then go back and take a detailed look at all your information you've put forward (in your post I'm quoting). As I understand it, the current purpose of our conversation is you proposing that only certain headphones should be used on the 5128 during the Target Creation & Listening test process - for finetuning the target curve.

 

[Mad_Economist]

So B&K should not be the standard until the research is done.

If you're aware of a larger set of data on human outer ear geometry to the eardrum and acoustic input impedance than Søren's paper, I'd love a link! I really liked their methodology for accounting for insertion depth variation on the probes.

If what people listened on is the same tonality as what was mixed on, it would sound better. That is the whole basis for the Harman Target.

That isn't necessarily true! It is not empirically proven that the playback system is compensated for directly by the production engineer, and in fact one of the citations Floyd has in his masterful Sound Reproduction (Børja 1977) actually explicitly shows that the in-room response difference between mixing rooms != the difference between recording spectra.

Yes it does. If you and they claim it is better, you have to show it is better. You can't just say it.

I think you two are getting a bit lost in the weeds of pedantry about what "better" means - it is objectively true that the 5128 matches human ear Z (including that of Gunnar Rasmussen and Per v. Brüel, on whom the 60318-4 coupler was based) better than the 60318-4 coupler. This means it is "better as a proxy for human ear Z". As a result it is very likely that with the correct model it is also at minimum equivalent and likely at least somewhat "better as an indicator of perceived frequency response of headphones by humans" - that work has yet to be done.

Except the rig Amir uses is compatible with the target and research.

As aforementioned, this is true to the same extent that the 5128, 4128, or HMSII.3 are compatible with the Harman research, effectively - there are real differences between the modified KB0070 Todd made and the KB501x pinnae used by those of us not named Todd Welti or Sean Olive, and the magnitude of those differences from the measurements on real humans isn't that different from the 5128.

A dogmatic reading of the Harman work - e.g. "you must have exactly the system used in the Harman research to produce meaningful projections of expected sound quality" - would thus imply that the KB501x is unsuitable. I don't endorse that reading, and IMO it's quite contrary to both what Sean set out to do and a reasonable read of the lit.

 

[Mad_Economist]

But there's more potential to feel exposed if you jump straight into a public conversation, I think far better for your initial discussions to be in private so you know where you stand, and then you're in a better position to do a public piece. It does feel like it could get quite personal though, but that's probably more likely to happen if you jump straight into a public video discussion.....I mean I can't imagine a public meltdown but people don't want to be put into a position where that can be a possibility, talking in private first is better.

I can see where you're coming from, and that definitely wouldn't be what I'd want - my goal here is to figure out something that makes Amir down to have a chat about this with me/us and hash out some of our differences, since text dialogue has...abjectly failed there.