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

[GaryH]

Spirited discussion here it seems - it's a long holiday weekend in the land of freedom, so you'll pardon me if I refrain from working too hard in my replies.


Re "pluralist nonsense" - glad we waited out Pride for that framing - we're referring consistently to Segmentation of Listeners Based on Their Preferred Headphone Sound Quality Profiles By Olive, Welti, and Khonsaripour here. In general, we are specifically talking about enacting what's highlighted here from Sean's conclusionView attachment 296502
In general, the goal here is not to "make headphones/IEMs look better", but to come up with something that better matches the subjective impressions of certain subgroups of listeners, based on the data that Sean has kindly provided. None of this is meant to indict the Harman target, and indeed the predictive power of the Harman target and Sean's statistical models is and will remain a canonical internal reference point for us - we are not looking to publish something that makes things muddier, except in the areas where people are overfitting.

People overfitting you say? I know that paper is the subjectivist's darling, but it really doesn't demonstrate what you all at headphones.com are hoping it does to use as a crutch for your pluralist FR narrative. As the classes in the paper were determined through cluster analysis of preference ratings given to a set of headphone frequency responses, the classes are partly dependent on the predefined FR selection of this set, which could well skew the results. As most headphones (at least at the time of the study) either had boosted (mid/upper) bass, or lacked bass (most open-backs), as seems apparent from the below graphs, it's unsurprising 3 classes of preference were found (the other being the large majority of Harman target lovers of course).

And the study was on headphone FRs, not IEMs, so any results cannot necessarily be assumed to hold for the latter and used as justification for non-Harman IE or a proliferation of in-ear targets/FRs. Disregarding these issues for now, let's take a closer look at the results:

As seen above (and Sean states in the very passage you quoted), the highest rated frequency response in both listener Class 1 ('Harman Target Lovers') and class 2 (supposedly 'More Bass is Better') is actually...the Harman target, with Class 3 (supposedly 'Less Bass is Better') rating it almost as highly as Class 1. So that's a vast majority superclass who most prefer the Harman target. And the highest rated headphone FR in Class 3? This, HP13 (Sony MDR-7506), which...follows Harman pretty well (the specific unit that they measured and so the FR under blind test):

Just a bit lacking in sub-bass and a bit more treble, but that's exactly the preference you'd expect from a class predominantly aged 50+, and so high-frequency hearing loss setting in. Crucially, this class also consists entirely of untrained listeners. We know from previous Harman research that older and untrained listeners are less reliable in their sound quality judgments, so it's quite possible they (or at least many of them) judged HP13 and the Harman target to be approximately equally preferable within the higher error margin of their judgements.

Finally, take a look again at Figure 3 above. The Harman target clearly has a universally positive rating across all classes i.e. all three classes of listener approximately agreed on its rating, and these ratings are high (which cannot be said for any of the headphone FRs). So, the Harman target does in fact suit everyone, and therefore should be the default curve targeted, after which users can if they feel the need make minor EQ adjustments to taste. This is the only way to break out of the circle of headphone confusion, and headphones.com's resistance to and moving away from Harman as a de facto standard, toward an untested target or proliferation thereof only does harm to this aim and the pursuit of better, more consistent headphone sound quality for everyone.

 

[Merkurio]

Right, so this one is an area where I think that there's been some bad messaging from us - in our defense, we expected to have a full-user-interactable measurement comparison tool ready by now, but hey, gotta plan for bad outcomes - and I want to make something a bit clearer: the target used on headphones.com derived from DF is not something that's being suggested as a most ideal response for headphones forever, or an overturning of the Harman work, or anything like that. It's literally a continuous slope applied to the DFHRTF, because the content team refused my initial pitch, which was only compensating to DF, and I didn't want to have any wonky little adjustments in there that would make it difficult for people to "undo" the preference adjustments. In the long term, my recommendation to Headphones.com has always been to use diffuse field alone for comparisons, with a "window" of the variations in preferred response that @Sean Olive has documented visible around the compensated trace - that remains the case.

Odds are that in the nearish future, we'll finally have our graph comparison tools operational, and people will be able to pick which of DF, DF with some preferential adjustments (probably user-set, as on @crinacle's site), and for 60318-4 measurements with GRAS pinnae Harman compensation they prefer. That has always been the endgame, although we've had a somewhat wiggly path to get there.

Ultimately, I've gotta take the blame here - I was really ardent that because of the sine illusion, we should move away from showing raw measurements of headphones and targets featuring HRTFs, because humans just plain objectively don't intuit the delta between said very well. The pushback I got was that compensating to DF alone would make people think that DF flat was the preferred response (which we have 25 years of data debunking), and compensating to Harman wouldn't work with the 5128 data. What some folks have done in that situation is trying to directly copy the Harman target over to the 5128, but given the differences of their pinnae, HRTFs, and acoustic Z this is not going to give people an accurate picture, and I'll die on that hill. Unfortunately, we couldn't get past the loggerheads between my refusal to use a more direct "copy" of Harman, and the content team not wanting to publish data with no preferential recommendations whatsoever.

The "split the baby" solution I proposed was that we just make the adjustment as simple as it possibly could be, minimizing the odds that we'd distort or mask some feature of the 5128, and making it as easy as possible for end users to remove that editorial adjustment both visually and if they chose to scrape the data. Some other options were floated - such as using the 1974 "Møller curve" from B&K, which Oratory does for his "optimum hifi" target - but ultimately we went with the simple slope under that logic. It was simply meant to represent the average tilt of the response which had been preferred in Olive, Welti, & McMullin 2013 vs. the flat in-room measurement, which is extremely close to the DFHRTF of the KEMAR. I've pushed back pretty hard against attempts to "fix up" the sloping so-called target since then specifically because I don't want people to think that this is some effort to propose a new way of doing things, and because I'd much rather push towards my original recommendation of normalizing straight-up DFHRTF compensation, which puts headphone frequency response in a psychoacoustically comparable space to in-room loudspeaker response (but without the annoyance of DI).

Hopefully in the very near future we'll have our data vis for content fixed up, and a lot of these concerns (and the combination of the slope with the DFHRTF in a single target) will be put to bed, and in the somewhat less near future we'll have some tools up on the site to let users play around with the kind of adjustments they'd prefer to make, similar to @crinacle. It's a work in progress, and the order of operations definitely got botched here, which is on me, but we are trying to make this stuff less opaque and more user friendly, even if it can look like the opposite sometimes.

I really appreciate your response, it definitely clears up many doubts I had and gives me (and surely to several people here) a clear perspective of the future you aim to achieve.

For now, I continue to be loyal to the measurements of the GRAS systems and their clone IEC derivatives, especially considering the established Harman target for that standard (which has become my de facto baseline for subsequent adjustments) and given the impossibility of using that same target on a 5128/4620 for the same obvious reasons you mentioned.

However, I will definitely be eagerly awaiting advancements in this subject, particularly in relation to creating a comparative tool for graphs and EQ, which is my most relevant use case.

In the end, despite the discordant tone in the last few pages, I believe we can all agree that it's better to have a debate about measurement standards than not having them. This speaks volumes about the growth and significance of the objective aspects of audio in the minds of more and more people. I think we can take that as a point of reconciliation and settle the matter with a vote of confidence, once it is clarified that there are no plans to replace GRAS system measurements in the near future and both can coexist perfectly.

Thank you as always, see ya' in the ASS (pun intended).

 

[amirm]

I have no idea of whether there is a statement I, or Headphones.com, or whomever you're indicting here can make to dispel this idea, but I'm going to try anyway: No, what I want people to think is that the 5128 is more accurate in its approximation of human ear impedance.

No reader of 5128 measurements understands what that means. Again, they are not looking for impedance of anything.

This is because the 5128 is more accurate in its approximation of human ear impedance.

Maybe. But again, that only has value if it translates into real life data predicting tonality of a headphone/IEM. This effort has been a failure so far.

Which, if the aim of reviewers was to measure wideband ear impedance, would be a real problem. However, while the input Z of the human ear does change with insertion depth for insert earphones (thus all this "propagation" business), that does not mean that it varies randomly, it's just about the modal resonances at high frequency shifting a bit. More accurate ear Z means a more accurate load, which, for cases where that matters, does mean somewhat more predictive measurements. It's the same as the argument for a 60318-4 coupler over a .4cc coupler, in fact, which long precedes demonstrable sound quality preference prediction differences between the two couplers.

That is not what the paper says:

"In the bare measurement results as presented a significant variability in impedance across subjects is noted, especially at frequencies above 3-4 kHz where resonances and anti-resonances occur. This variability is mainly due to a variable path length between the measurement planes in the ear canal and the ear drum, when measuring the impedance from subject to subject. It is difficult in practice to maintain the same path lengths, or insertion depth, across all subjects, partly due to the large intra-subject variation in ear canal geometries, where some ear canals are quite narrow with limited space to fit the waveguide tubes of the transmitter and receiver. "

You also did not address the major point I made that the study could not capture reliable measurements of almost 1/3 of the study participants. In other words, the research protocol was not sufficient to generate proper results putting in doubt your claim that the research was able to determine the impedance of human ear. It is entirely possible that another research would arrive at a different mean impedance and deviation.

 

[Robbo99999]

I have no problem whatsoever with changing the standard of measurements to one that is more accurate and more in line with the purpose for which it was developed (a better representation of human hearing), after all that is the purpose of improving in all fields of knowledge and acoustics were not to be outdone.

Nevertheless, and from my narrow perspective as a mere aficionado (who values the importance of the graphs as a mere approximation of what will finally reach my eardrum), I consider that the robustness of the targets developed based on GRAS systems outweighs (for now) the added accuracy that B&K systems can provide, at least until they're not on equal terms to be used for EQ purposes.

In that sense, I think it wouldn't hurt to have a disclaimer telling people that the measurements done using GRAS systems are (to date) more robust to use for EQ headphones, while the B&K ones are a more accurate (state-of-the-art) systems to better represent the FR of the same headphone (although I insist, without a familiar target like the Harman that serves as a compensation, we are in the same blind spot regarding to the increased accuracy it provides).

Just my opinion as someone who does not see things from the same technical stature as you do guys.

Disclaimer idea is quite good, afterall you don't really want to allow for the possibility of misleading your audience, but that'll be up to headphones.com as to how they approach it, and since your post Mad_Economist has said it's something they might consider. Obviously B&K 5128 is currently not as valid as GRAS for EQ'ing to a known standard (in other words Harman Curve).....and it will take a lot of work on the 5128 to come up with a target curve that can carry the same weight as GRAS Harman. (And the fact that B&K 5128 is more representative of an average human ear doesn't sway that conclusion, not until rigorous enough work has been done to create a valid Target Curve that would hopefully be a proven preference curve created as a result of their study).

But, I've said it before, I'm happy to see Resolve & Co seeing what they can come up with re B&K 5128. What the forum going & Youtube population of headphone.com users think of 5128 in terms of validity vs GRAS I don't know as I don't really participate or read/watch stuff over there, but hopefully it's pretty balanced & informed (EDIT: given the huge breadth of user types that use Youtube, then it's probably not that balanced - it would require concerted effort in the Youtube reviews (probably briefly referred to in a larger proportion of the reviews) to highlight the differences or instead main take away point(s)).

 

[Mad_Economist]

No reader of 5128 measurements understands what that means. Again, they are not looking for impedance of anything.

Maybe. But again, that only has value if it translates into real life data predicting tonality of a headphone/IEM. This effort has been a failure so far.

This is essentially an equivalent argument to that which could be made against the use of head and torso simulators over omnidirection microphone capsules for measuring headphones in 2013. It would have been about equally uncompelling to me then as well - the Harman work is exceedingly good science and we are privileged that so much of @Sean Olive's work is publicly available, but the case for loading systems with unknown and difficult to quantify Zout with accurate load Z is self-demonstrating.

That is not what the paper says:

"In the bare measurement results as presented a significant variability in impedance across subjects is noted, especially at frequencies above 3-4 kHz where resonances and anti-resonances occur. This variability is mainly due to a variable path length between the measurement planes in the ear canal and the ear drum, when measuring the impedance from subject to subject. It is difficult in practice to maintain the same path lengths, or insertion depth, across all subjects, partly due to the large intra-subject variation in ear canal geometries, where some ear canals are quite narrow with limited space to fit the waveguide tubes of the transmitter and receiver. "

...did you actually read the paper? The commanding majority of it is specifically about accounting for variations in path length mathematically. That's why the phrase "bare measurements" is used there. That's why there's a "propagated" column and as you can see, the propagation works, variation falls substantially when the effective volume length is matched between the subjects.

You will also note that there are no apparent cases of a subject falling towards the bottom of the Z range in the mid range and the top in the bass, as occurs with the 60318-4

This, again, is not to say that the 60318-4 coupler is useless - it clearly is not - but rather that it demonstrably behaves in ways that are further from the typical behavior of the human ear than the ITU-T P57 4.3 ear.

You also did not address the major point I made that the study could not capture reliable measurements of almost 1/3 of the study participants. In other words, the research protocol was not sufficient to generate proper results putting in doubt your claim that the research was able to determine the impedance of human ear. It is entirely possible that another research would arrive at a different mean impedance and deviation.

It is difficult for me to believe this statement is in good faith. Removing subjects where clear confounding effects prevented reliable measurements is a necessary part of any acoustic measurements on human subjects, and particularly with measurements using constant displacement probes, even tiny leaks in the earmould would create substantial errors in low frequency Z. You will see this in pretty much any good study involving probe mics in human canals because small variations can have large effects here, and avoiding swamping your results in noise requires some data processing, including saying "we can't fit this particular canal with our transmitter-mic earmold, we'll have to exclude this subject from the test".

Your alternative here would be either vastly increasing the time required - which smacks strongly of "why didn't Olive just have 5000 subjects in each of his listening tests?" type of unrealism in terms of expectations of study design - or get completely non-representative results because you left in a bunch of erroneous measurements. I mean hell, this was an issue with the IEC711 research as well, and that was only the two experimenters' ears!

 

[Pe8er]

"why didn't Olive just have 5000 subjects in each of his listening tests?" type of unrealism in terms of expectations of study design

Can you explain why is that unrealistic?

 

[Mad_Economist]

Can you explain why is that unrealistic?

For Olive to have 5000 people in his listening tests, or for Søren to leave out some people from his MRI cohort? In both cases, the answer is fundamentally "resources are finite and limited, and science has to make do". In the case of the paper on ear Z, the only reason that the extra 12 people are mentioned at length is that their ear scans were still part of the design of the ear simulator, even though for various reasons fitting the Z measurement probes to their ears was not viable - I would postulate that leakage would be the dominant problem (and the most time-intensive to correct, since you would need to fabricate new earmoulds), but based on their commentary some of the moulds may also simply not have had space for the probe microphone and transmitter arrangement (and eyeballing the 3D ear models and seeing which ones didn't make it, several do seem to have rather narrow canals, although not all the narrow canals were omitted).

 

[amirm]

...did you actually read the paper? The commanding majority of it is specifically about accounting for variations in path length mathematically. That's why the phrase "bare measurements" is used there. That's why there's a "propagated" column and as you can see, the propagation works, variation falls substantially when the effective volume length is matched between the subjects.

Of course I read it. Who do you think I am? I quoted that specifically to point out you as reviewers have no prayer of achieving the same loading impedance as in the experiments.

It is difficult for me to believe this statement is in good faith. Removing subjects where clear confounding effects prevented reliable measurements is a necessary part of any acoustic measurements on human subjects, and particularly with measurements using constant displacement probes, even tiny leaks in the earmould would create substantial errors in low frequency Z.

No it isn't. It is a significant failure in measuring the impedance of those subjects. The issue was also not error but unreliability of repeated measurements pointing to experimental error. This is the core topic we have been discussing from day one related to 5128. You want to sweep it under the rug as it is to be expected???

 

[amirm]

I mean hell, this was an issue with the IEC711 research as well, and that was only the two experimenters' ears!

We are NOT in that hell because we have something better: correlated measurements with listening test results. You are working backward with 5128 by trying to say it is good enough absence of such listening test results. Well, you have a single experiment with nearly 30% failure to measure a critical parameter. You can't sugarcoat this with whataboutism.

 

[amirm]

Let me state it more clear: the very experiment to determine the average impedance of human subjects failed to produce reliable results for 1/3 of the subjects. That impedance was then implemented in an ear simulator in the form of B&K 5128 which just the same is showing high variability. This is no way to get "accuracy." Yes, it is an attempt but in dire need of independent verification.

 

[Mad_Economist]

I quoted that specifically to point out you as reviewers have no prayer of achieving the same loading impedance as in the experiments.

Okay, so perhaps we should take a step back here and ask if you understand why ear load Z changes and what changes it - the insertion depth/volumetric modes, of course, shift as a product of the actual length of the canal (terminated at the IEM/earmold/etc on one end, and the drum on the other, or open in the case of something that doesn't occlude the canal). The low frequency Z changes in the presence of leaks. In both of these respects, both ears and the 4.3 style sims behave the same - with 60318-4s, you have a maximum insertion depth at the grille protecting the primary volume, not that this matters overly much.

Regardless, while you won't get the same insertion depth that Søren found, that does not change that for a given insertion depth and leakage, the 4.3/Type 4620 ears are more accurate to human ears. And getting something rather consistent with human experience in both respects is rather important, since those factors meaningfully impact the sound pressure at the eardrum.

No it isn't. It is a significant failure in measuring the impedance of those subjects. The issue was also not error but unreliability of repeated measurements pointing to experimental error. This is the core topic we have been discussing from day one related to 5128. You want to sweep it under the rug as it is to be expected???

There is nothing under the rug here! We have a geometric model based on 44 subjects, and an impedance average based on 32, the reasons for which are explained clearly in the paper! From a mechanism point of view, what source of error are you even suggesting would occur here that would bring the other measured Z there into disrepute? Quantifying acoustic loading of a small volume is not exactly a controversial area of physics, and if substantial leakage or insertion depth issues had occurred, they would be clearly visible in the data (because the effective volume size/Z would rise much faster at LF than it in fact does, and the length modes would not be at the correct frequencies).

We are NOT in that hell because we have something better: correlated measurements with listening test results.

I mean, if you want to be this stringent, as I have pointed out before, no, "we" don't - Sean Olive, Todd Welti, and Harman International do, because they have Todd's custom pinna, which was used in the Harman research. The KB5010/5011 you use and the Type 4620 pinnae of the 5128 actually have very similar difference to the Welti pinna in the treble.

Generally, I'm not indicting measurements made on KB501x systems on this basis, because I'm willing to look at the mechanisms that underlie those correlations and say "is there a reasonable cause for concern that this other, humanlike system is going to predict eardrum SPL? And, if not, is this a major issue?" If you feel differently, perhaps you should send Sean an email and request some copies of his ears.

Well, you have a single experiment with nearly 30% failure to measure a critical parameter. [...] the very experiment to determine the average impedance of human subjects failed to produce reliable results for 1/3 of the subjects.

If the study was "a novel design for earmolds enabling zero leakage in situ in human ears", I'd agree. However, it wasn't, and the fact that some ear molds could not be constructed without parasitic leakage or other effects has very little bearing on the paper's results - indeed, I'm really straining here to see how you see any problem with this. Again, the purpose of this paper was 1, to document the average canal geometry, and 2, to document the average Z, and it's completely unclear to me why you feel that excluding some of the sample population inherently makes 2 compromised.

That impedance was then implemented in an ear simulator in the form of B&K 5128 which just the same is showing high variability

High variability in what respect? I'm not aware of any QC issues reported with the Type 4620s, and their results appear quite consistent across users when operator effects (e.g. placement and coupling error) are set aside.

Edit:

You can't sugarcoat this with whataboutism.

I really think you need to take a long look at this statement and ponder whether I, who have stridently argued in favour of 60318-4 measurements for years, and continue to, am attempting to "whatabout" one of my favourite works, by authors who are both seminal to the field and personal heroes of mine, to defame a coupler which I personally use, will continue to use, and continue to advise people to use, or if you are reading some bizarre agenda into my posts.

 

[Tallulah]

You can see there's an 8khz peak [...] The HD600 doesn't actually have an audible ~6dB peak there, does it?

I've been trying Resolve's equalization for the HD 600... I can confirm (using a tone generator) there's no peak at 8.1 kHz to my ears. In fact, the response there is reduced for me. Oratory1990's measurements and correction are way more representative to my hearing. I also had to use the -2 dB filter at 3 kHz with Resolve's correction, which only made it more similar to Oratory's correction for Harman target.

 

[Robbo99999]

Quick post: I'm not able to follow every single "in & out" in @amirm 's & @Mad_Economist 's discussion as some of it is above my current pay grade so to speak, but at least we're not being overtly rude to each other, so I'm OK to see the discussion unfolding.

 

[GaryH]

based on their commentary some of the moulds may also simply not have had space for the probe microphone and transmitter arrangement (and eyeballing the 3D ear models and seeing which ones didn't make it, several do seem to have rather narrow canals, although not all the narrow canals were omitted)

So the acoustic impedance measurements/calculations are likely skewed to people with larger ear canals then, invalidating the results and comparison with the 711 coupler, the supposed significant differences with which are the whole basis you're using to claim the 5128 is superior with a more accurate to human acoustic impedance. Meaning this conclusion from the paper:

At the lowest frequencies the level of the IEC711 coupler is slightly higher, probably indicating a slightly too low equivalent volume of the IEC711 coupler

could well be erroneous, with the actual reason for the difference being the mean human ear canal volume of the 32 subjects used in determining the impedance is too high due to the exclusion of subjects with smaller canals that could not be reliably measured.

 

[Robbo99999]

So the acoustic impedance measurements/calculations are likely skewed to people with larger ear canals then, invalidating the results and comparison with the 711 coupler, the supposed significant differences with which are the whole basis you're using to claim the 5128 is superior with a more accurate to human acoustic impedance. Meaning this conclusion from the paper:

could well be erroneous, with the actual reason for the difference being the mean human ear canal volume of the 32 subjects used in determining the impedance is too high due to the exclusion of subjects with smaller canals that could not be reliably measured.

The bit I bolded in your post - for me this crossed my mind as I read his post too.

 

[amirm]

Regardless, while you won't get the same insertion depth that Søren found, that does not change that for a given insertion depth and leakage, the 4.3/Type 4620 ears are more accurate to human ears.

You have eliminated important caveats in repeating that marketing talking point. That assertion is based on a limited study where nearly 1/3 of the subjects were excluded from the project because their protocol failed in capturing a reliable measurement. The study hasn't been repeated and no pragmatic results have come out of any such increase in resolution. Any fair reporting of this topic would have hit on these weaknesses but we don't see that. All we see is the bolded marketing line.

The whole thing is a house of cards as far as applicability due to the listener base. Let's review the graph again:

Look at those incredibly tall error bars (light green) that are even falling out of the bottom of the graph at just 6 kHz! The fact that you can compute a mean out of them doesn't let you out of the quandary that you are leaving a ton of listeners behind. So what that the dip at 14 kHz shouldn't be as deep as it is on a stock 711 fixture. You can't quantify much in that region anyway.

For the purposes of standardization and having something that is repeatable between test labs, these averages work. This is why IEC listening room exists for experiments in room, not because it has much science behind its specifications. That is not what we want out of a fixture. We are not running lab experiments that we want to publish and have others replicate. We want to predict listener preference and tonality of response and for that, the findings of this study shows how bad the situation is due to highly variable aspects of human ear canals. It is a miracle that Harman managed to make some sense out of these measurements after much work and refinement.

What this means is that none of these measurements at higher frequencies have much measure of accuracy. To claim that a match to "human Z" has been found where the very experiment shows human impedance is all over the place means your statement is just pure marketing.

Now, for R&D purposes where you keep the fixture constant and you iterate a design, you can clearly tease out what is changing. This is how B&K is selling the 5128. That is the reason they have made no attempt in creating a preference curve for it. Trying grab the fixture against that and run with claims of more accuracy for reviewing headphones to give broad range of listeners a sense of how they sound, is just irresponsible.

There is no question in my mind that these fixtures are purchased in a competitive race to what other reviewers are doing with nary a thought as to whether they are helpful in answering any useful questions after you spend some $40,000 on it. It is height of subjectivism to spend money first and figure out the answer second. Of all people, I am the one that wants the best there is when it comes to measurements as to reduce the amount of criticism over them. But here we are folks jumping on a new fixture where questions are front and center and answers are just repeating marketing lines.

 

[amirm]

I really think you need to take a long look at this statement and ponder whether I, who have stridently argued in favour of 60318-4 measurements for years, and continue to, am attempting to "whatabout" one of my favourite works, by authors who are both seminal to the field and personal heroes of mine, to defame a coupler which I personally use, will continue to use, and continue to advise people to use, or if you are reading some bizarre agenda into my posts.

The only thing I need to look at is that you have taken a consulting position to headphones.com and put forward as the expert witness on why they have purchased the 5128. If this is wrong you need to come out clearly and properly state that purchasing a 5128 for headphone review is a mistake. If you don't, and keep defending it by throwing papers at people while leaving out key detail, then I am going to call you on both fronts.

 

[Mad_Economist]

You have eliminated important caveats in repeating that marketing talking point. That assertion is based on a limited study where nearly 1/3 of the subjects were excluded from the project because their protocol failed in capturing a reliable measurement.

I will say this again, if you understand the actual causes of variation in ear Z, you will see that it is by no means likely that excluding subjects has biased the trends of the ear Z here - there are a finite number of mechanisms of error, and their traits are pretty obvious in the data. Søren and co have excluded the data showing these predictable error mechanisms - again, I'd allege that low-frequency leak, visible as a higher order rise in the equivalent volume dimensions, is probably the major cause, but I didn't work on this paper, so that's conjecture

All studies are of course limited - this may be said of pretty much anything under the sun. This study is, however, the more comprehensive study of human ear impedance, and it accounts for the largest range of confounding variables - it is "the state of the art", such as it presently exists, unless I've missed a more recent publication (which I'd be very eager to read if I have - please link it).

Look at those incredibly tall error bars (light green) that are even falling out of the bottom of the graph at just 6 kHz! The fact that you can compute a mean out of them doesn't let you out of the quandary that you are leaving a ton of listeners behind. So what that the dip at 14 kHz shouldn't be as deep as it is on a stock 711 fixture. You can't quantify much in that region anyway.

The variation at 6khz is pretty evidently because that node wasn't the alignment point used in the paper

however, that's also a product of the volume length, which, as you note, will inevitably vary both with insertion earphones (because of fit variation) and with open canals (because canal length varies). It's helpful to look at the causation of variation when we consider if it's problematic or not.

We want to predict listener preference and tonality of response and for that, the findings of this study shows how bad the situation is due to highly variable aspects of human ear canals. It is a miracle that Harman managed to make some sense out of these measurements after much work and refinement.

I'm departing this conversation, but rhetorically I have to ask, what is the sense and work you see in the Harman work, if you, as previously discussed, completely discount the predictive models? What differentiates Olive's work from ex. Fleischmann or Lorho? Is it simply the volume of papers and listeners? It would seem that you treat (some unspecified subset of, not including the bits on predictive modeling) Sean's work differently than other works in this field in that respect, but I'm not entirely clear on what it is that makes you more prone to doubt some papers than others.

What this means is that none of these measurements at higher frequencies have much measure of accuracy. To claim that a match to "human Z" has been found where the very experiment shows human impedance is all over the place means your statement is just pure marketing.

This seems like it misunderstands the fundamental cause of variation in Z in this band, which is kind of a recurring theme here, but deterministic variations are not the same thing as random noise.

There is no question in my mind that these fixtures are purchased in a competitive race to what other reviewers are doing with nary a thought as to whether they are helpful in answering any useful questions after you spend some $40,000 on it. It is height of subjectivism to spend money first and figure out the answer second. Of all people, I am the one that wants the best there is when it comes to measurements as to reduce the amount of criticism over them. But here we are folks jumping on a new fixture where questions are front and center and answers are just repeating marketing lines.

I suppose I can't stop you from adopting this framing - and, certainly, if one wishes to apply your own style of "painting bad motives onto your interlocutor", it's quite beneficial to you, as a non-5128-owner, to adopt this stance - but the very concept of "the best there is" regarding HATS is a bit misleading to begin with, and I think that your anchoring to that premise is part of why you have a weird complex about people talking about the technical merits of the 5128. There are legitimately meaningful issues with all artificial ears and heads, documented across the lit - including, recently, by Sean's comparison with blocked meatus microphones - when it comes to predicting in situ response on human heads. No one head is a canonical truth of the subjective experience every user will have with a headphone, and indeed, it's very likely that we must develop metrics that account to some degree for variation here (in the same way that DI predicts, to an extent, sensitivity to room variation) if we want to improve in the long run on Sean's models (good though their fit generally is).

The only thing I need to look at is that you have taken a consulting position to headphones.com and put forward as the expert witness on why they have purchased the 5128. If this is wrong you need to come out clearly and properly state that purchasing a 5128 for headphone review is a mistake. If you don't, and keep defending it by throwing papers at people while leaving out key detail, then I am going to call you on both fronts.

Ironically, the commanding majority of the consulting I do for Headphones.com is helping them interface with some Chinese manufacturers I know - I don't think I was even in any of the content they put out regarding the 5128, and I'm posting here on my own time (indeed, I got a message at some point along this thread to the effect of "you know that we're not paying you to argue on the internet, right?") and with my own motivations.

However no, I do not need to state that purchasing the 5128 for headphone review is a mistake, and no amount of hamhanded attempts to force that dichotomy will do so, because I strongly supported buying the 5128, and am glad that firms like RTings have added it as well. What would have been a mistake is saying "all 60318-4 measurements are fundamentally useless, ignore all of them, we're throwing away our 43AGs, we'll only do 5128 measurements from now on", and I'll happily condemn the hypothetical alternative timeline nega-@Resolve that said that in universe 616. However here, in the reality that actually occurred, adding additional data from an additional pinna alone would be justification to purchase a second HATS, which was indeed part of my specific argument for buying it, with the added benefit of giving us a picture of the behavior of high acoustic Z designs with more accurate loading.

This strange premise that there is simply a "best ear" which should be used to the exclusion of all others is, frankly, an infantile assumption that everyone is buying these tools in an attempt to defame all their "competition" - you may be looking for a reason to say that other people's data is untrustworthy, but that isn't the motivation of everyone who buys a new metrological tool. Your repeated insinuations and outright statements that I'm some sort of paid shill for holding a position which I've demonstrably held since before I got one red dime from Headphones.com (and, sadly, I have yet to be paid at all by Hottinger Brüel & Kjær) and attempts to pin on me, @Resolve, and everyone who says something positive about the 5128 some sort of narrative that 60318-4 measurements are useless or should be disregarded out of hand, say everything about the degree of insecurity and incivility with which you're approaching this.

I suspect you are having some sort of protracted, imaginary argument with Jude Mansilla or someone else who you dislike who was very pro the 5128 - I don't particularly care, I'm not going to be the stress ball for it. I have attempted to explain to you, at length, the real but not disqualifying to extant systems technical merits of the 5128, the relative issues which exist with all headphone metrology, and why I'm quite pleased that both KB501x and Type 4620 data (and, hopefully soon, data from the new HMS, and its distinct pinnae) is increasingly available, and my hopes for where we can go with data from multiple pinnae, and with a more realistic ear load for in-ear headphones (particularly, ironically, at low frequency - I feel like people keep ignoring that, but that's actually where the 5128 is meaningfully more interesting IMO). I've attempted to do this civilly, and I haven't accused you of financial motivations, petty vainglory, or any of the other immediate reactions that your inane and bad-faith responses provoked, because we aren't even on opposite sides here - this is a "conflict" that you have manufactured between the 43AG/45CA and the 5128, and I have refused - and continue to refuse - to make it real, because it isn't, at least for me, for Headphones.com, for Crinacle, and for every other interested party I'm personally connected to who has worked with the 5128 which I am personally familiar with (Linus Media Group, HypeTheSonics, RTings).

You have created a pointless argument here and made it pointlessly uncivil, and produced a wealth of ammunition for everyone out there who wants to paint the pursuit of scientific audio as venal and self-interested here. I have a distinct flashback to your past conflicts with Erin, Arny (RIP), and heaven knows how many others. I am no longer feeling like taking the high road, so I will be taking my leave instead - you will, naturally, have the last word, as is your wont.

 

[peniku8]

I've been trying Resolve's equalization for the HD 600... I can confirm (using a tone generator) there's no peak at 8.1 kHz to my ears. In fact, the response there is reduced for me. Oratory1990's measurements and correction are way more representative to my hearing. I also had to use the -2 dB filter at 3 kHz with Resolve's correction, which only made it more similar to Oratory's correction for Harman target.
View attachment 297201

I'm not 100% sure about the HD600, but my HD650s are quite different from the ones Oratory measured (or the average he made?), because they were terribly sibilant and I had to EQ away a lot of top end. High unit variation seems to be an issue with Sennheiser, since the IEMs I had also didn't sound right with any of the EQs I found from Crinacle or Oratory. Curiously, the HD800 unit variation was really good.
@oratory1990 how many HD600s fed your data? Happen to have a unit variation graph at hand?

 

[amirm]

I suspect you are having some sort of protracted, imaginary argument with Jude Mansilla or someone else who you dislike who was very pro the 5128 - I don't particularly care, I'm not going to be the stress ball for it.

Nonsense. I am not driven by emotion. I stated what worries me: that right when we were getting close to having the industry to converge around a modified 711 coupler, here comes the 5128 parade confusing the matter, leaving the industry in a fog. Witness the latest headphones from DCA and their compliance with 711. See Crin's IEMs. We were getting there with Harman target on modified 711 before you all started to spit in the soup.

In that context, I don't care how you feel about my interactions with you. I am going be super blunt calling out a commercially motivated effort that is interfering with what we as consumers badly need, a unified standard to rally around. Not a darn thing was broken as far as instrumentation to go and mess with that. What is broken is that so many companies are still producing random frequency response headphones/IEMs. Your effort lacks wisdom in this regard and is what has gotten us here. This makes me very sad.

What a screwed up industry this headphone business is....