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

[Chagall]

It certainly is not! The anechoic on-axis response of a speaker is informative.
Now what's the headphones' equivalent to a speaker's anechoic frequency response? For the speaker we removed the room, so for headphones we need to remove the listener's (or test fixture's) "head", as you have correctly identified previously:

So what you suggest a simple pressure field measurement of the headphones. No pinnae, no ear canal, no possible variation, just raw data.
It's an interesting proposal and I wouldn't be surprised if headphone manufacturers actually take measurements like this at least somewhere in their process (quite possibly for QC because that doesn't require an expensive HATS), but I'm not sure how informative that data would be for us consumers.

In Amir's analogy:
- anechoic speaker response measurement = headphone response measurement on a single standardized rig
- room modes with speakers = variations caused by real heads
He’s saying we don’t discard anechoic speaker measurements just because the frequency response will change in actual rooms.

The problem, as I see it...is that a in-room measurements are relatively easy and inexpensive to perform, $100 UMIK and REW software, anyone can measure their in-room frequency response.

However, measuring headphone response in situ (on your own head) isn’t nearly as straightforward. If it were, we wouldn’t need to rely solely on standardized rigs - they’d still be useful for comparison, but everyone could also measure their own headphones individually using reference headphones as a target.

 

[MayaTlab]

In Amir's analogy:
- anechoic speaker response measurement = headphone response measurement on a single standardized rig
- room modes with speakers = variations caused by real heads
He’s saying we don’t discard anechoic speaker measurements just because the frequency response will change in actual rooms.

The problem, as I see it...is that a in-room measurements are relatively easy and inexpensive to perform, $100 UMIK and REW software, anyone can measure their in-room frequency response.

However, measuring headphone response in situ (on your own head) isn’t nearly as straightforward. If it were, we wouldn’t need to rely solely on standardized rigs - they’d still be useful for comparison, but everyone could also measure their own headphones individually using reference headphones as a target.

It’s a false equivalence anyway.

 

[OHtaru]

In Amir's analogy:
- anechoic speaker response measurement = headphone response measurement on a single standardized rig
- room modes with speakers = variations caused by real heads
He’s saying we don’t discard anechoic speaker measurements just because the frequency response will change in actual rooms.

The problem, as I see it...is that a in-room measurements are relatively easy and inexpensive to perform, $100 UMIK and REW software, anyone can measure their in-room frequency response.

However, measuring headphone response in situ (on your own head) isn’t nearly as straightforward. If it were, we wouldn’t need to rely solely on standardized rigs - they’d still be useful for comparison, but everyone could also measure their own headphones individually using reference headphones as a target.

Not the same. Anechoic chamber means no room. Standardized headphone test fixture means ‘a standardized room’.

 

[glomple]

Harman published an article recently on the subject : https://aes2.org/publications/elibrary-page/?id=22943

Well seeing as the thread managed to turn into an argument about test fixtures, figures 9 and 10 of this paper are really interesting, here is a direct quote as well:

Finally, it is worth mentioning that while the Type
4.3 artificial ear does correspond to an average of
human ears, the Type 2 artificial ear seems to be an
outlier. It has only 2 closed ear canal resonances up
to 20 kHz and the small ear canal volume results in a
higher SPL than those in subjects’ ears as seen in Fig.10

So at least for the JBL Live Beam 3, the BK5128 seems to behave more like the real thing compared to the GRAS. And something that has been bugging me, if variability between measurement equipment and also between real human ears is so massive, for any standardized preference target you'd need a massive amount of listeners and a significant amount of targets to give them to listen. Preferably quite a few output devices as well. Without that any standard is honestly useless. But if you do have that well congrats you will probably get to pick whatever standard you want to show your conclusions on if you decide to publish. I do wonder how much proprietary research big companies like Apple or Sony are sitting on on this topic

 

[solderdude]

Without that any standard is honestly useless

It isn't.
A standard is supposed to be very well described as well as tolerances it might have.
This is so that different testers using the same setup/fixture and target can compare results.
The results are all made using the same standard so comparable.

Standards differ and fixtures (incl pinna and coupler) differ so the results will differ.

People's ears differ from standards and preference differs (targets).

The idea is to make a fixture to be 'close to an average ear' but take a look at ears and research in earcanal shapes as well as look at EDRP measurements.
They all can differ substantially from standards.

So standards are not useless they are valuable. What standards don't represent are people's ears (and perception) but chances are some ears are quite close to those standards as they are kind of modeled after some average of a number of ears.

 

[markanini]

You can't really divorce a target from the validation process behind it. This is why Harman OE 2018+GRAS 45CA remains the gold standard for how. Tidbits of data and marketing favorable to BK5128 as a human head simulator can't make up for that, especially when the data isn't consistently favorable to BK5128:

https://twitter.com/x/status/1608910510451458050

 

[glomple]

You can't really divorce a target from the validation process behind it. This is why Harman OE 2018+GRAS 45CA remains the gold standard for how. Tidbits of data and marketing favorable to BK5128 as a human head simulator can't make up for that, especially when the data isn't consistently favorable to BK5128:

https://twitter.com/x/status/1608910510451458050

That's very fair. The bigger problem I see is with TWS, which have come to completely dominate the audio market overall, but public research seems to be missing completely. So companies are either super secretive or they're winging it. I am not sure how applicable previous data is for them considering they typically have a shallower fit than IEMs and they are also used with ANC. So while the research was trying to catch up to what people would like listening to, the market ran away again.

 

[listener650]

You can't really divorce a target from the validation process behind it. This is why Harman OE 2018+GRAS 45CA remains the gold standard for how. Tidbits of data and marketing favorable to BK5128 as a human head simulator can't make up for that, especially when the data isn't consistently favorable to BK5128:

https://twitter.com/x/status/1608910510451458050

Honestly there are parts of this sentiment I agree with; I consistently find the 5128's results with over-ear headphones over 4 kHz or so to be... puzzling, to say the least.

I believe this is largely due to its pinna geometry being quite different from my own, as well as that—per Sean's data you shared—the pinnae of all HATS fixtures (including the GRAS KB5000 series ear) overestimate treble HpTFs between 5-10 kHz vs. human measurements... except interestingly, the Welti ear, which underestimates treble a bit.

For that reason, I really don't think there's any argument for 5128 over the GRAS KB5000 series over 4 kHz for over ear headphones, based on its pinna geometry alone... but of course, the pinna is not really what makes 5128 unique. It's the ear canal and ear drum.

Pinnae were the only part that the headphones were interacting with in that investigation because they used blocked ear canal measurements, but this investigation of Sean's was still incredibly important. It was, before the new study in the OP, among the first indications we had from Sean himself that using the KB5000 pinnae with the 2018 Harman Target was ill-advised because the pinnae produce meaningfully different HpTFs from the Welti ear (obviously they both use the same middle-ear simulator, so the pinnae/ear canal entrance is the source of the disconnect here).

Of course, we now have the paper in the OP that takes this a step further with open ear canal HpTF measurements and a variety of headphone/ear load interactions, indicating that the truth (frequency response measured on humans) is a much more varied story than the GRAS system or the B&K system alone will tell. The question then becomes: what is the best way to capture this information about a headphone for an audience who is reading/watching a review? I think we'll all have different ideas on that, but I assume we all probably agree with Sean that it's a question worth answering.

 

[listener650]

That's very fair. The bigger problem I see is with TWS, which have come to completely dominate the audio market overall, but public research seems to be missing completely. So companies are either super secretive or they're winging it. I am not sure how applicable previous data is for them considering they typically have a shallower fit than IEMs and they are also used with ANC. So while the research was trying to catch up to what people would like listening to, the market ran away again.

There is actually some public research being published by Harman that is useful not necessarily for characterizing what TWS earphones are doing, but what problems the really good ones—like the Apple AirPods Pro 2 & 3 and Bose QC Ultra earphones—are trying their best to solve. Namely, HpTF variation across human subjects when earphones are placed in their ears.

Check out 2023 - In-Ear Headphones on Ear Canal Simulator vs Real Human Ear Geometries- Quantifying the Differences with Simulations and 2025 - Estimating the Sound Pressure at the Eardrum and Ear Canal Transfer Function with In-Ear Headphones, both are public access and give really neat insight as to the issues that earphones will have when placed in human ears (and gives an inkling as to why I think eg. AirPods Pro 2/3 regularizing frequency response across individuals in-situ under 4-5 kHz is such a good idea).

 

[amirm]

Of course, we now have the paper in the OP that takes this a step further with open ear canal HpTF measurements and a variety of headphone/ear load interactions, indicating that the truth (frequency response measured on humans) is a much more varied story than the GRAS system or the B&K system alone will tell. The question then becomes: what is the best way to capture this information about headphone for an audience who is reading/watching a review? I think we'll all have different ideas on that, but I assume we all probably agree with Sean that it's a question worth answering.

In my reviews, I note how difficult/variable the measurements were on the fixture. If they are that finnicky there, it is likely to be just as much or more so on humans.

Before wondering how to deal with this, you have to ascertain that there is an answer. There are numerous variables here. How do you deal with pad wear for example? Or the very different ways people wear headphones in every instance?

My answer is that you just need to be comfortable with how far we have come. Countless enthusiasts are happy with the way measurements with EQ guide them to excellent sound. I know I am happy. Very happy. Learn to live with ambiguity. Not everything lends itself to black and white answers as my doctor said of his confidence in his diagnostic of me.

 

[amirm]

Now this is just a ridiculous analogy, you know it isn't in any way related to listeners message and yet still post it. Not trying to come across as rude... it just doesn't help fix your image here.

You are being rude. But the thing that gets me is posting an information-free post in a thread that is highly technical. And ignoring much of what I have been posting here.

The analogy is as perfect as it comes. Both speakers and headphones are used to transmit what is recorded to your ears. Research gives us measurements/tools/understanding to characterize likability of their sounds to fair degree but cannot provide assurance. My rule is that objective speaker measurements gets us to 70 to 80% and headphones, 60 to70%. In that regard, the two fields are identical in leaving unanswered questions. That doesn't mean in any way or shape that we should not care about the targets they propose, or that there is some kind of answer to the unknown.

Further, as I have repeatedly said, we need a standard here. Saying the target is of little value because there is variability leads us to terrible decisions like polluting the objective waters with yet another test fixture and one or more made up targets for it. Again, same is true of speakers but fortunately there, CEDIA led the effort to standardize Harman's research, making it concise and practicable. Same has not happened in headphones.

The variability for the listener has a perfect solution: use EQ. Adopt the filter set I or others publish for a quick answer. And modify if needed to taste. Maybe this can be automated at some point but it is not a big deal right now (except in some platforms like iOS). Get a headphone that is close enough to target as to not have EQ be mandatory for the exceptions.

Really, the work that Harman has done for headphones and speakers is a gift. It takes us out of the wild jungle of anything goes. Don't keep criticizing it and throw solutions at it without thinking it through and spending some energy in proper research. Kudos to Dan Clark who helped with the research in OP to validate the results of his 5128. The rest, sitting there just serve to be annoying -- just like your post.

 

[FTB]

My very modest and individual two cents are that every headphones or IEMs I tried which have good compliance with the Harman target on GRAS, are good to my (subjective) tastes.

I can also write that I don't lik the ones which don't fit the curve.

I don't have the same experience with 5128 for now.

 

[OHtaru]

My very modest and individual two cents are that every headphones or IEMs I tried which have good compliance with the Harman target on GRAS, are good to my (subjective) tastes.

I can also write that I don't lik the ones which don't fit the curve.

I don't have the same experience with 5128 for now.

Interestingly, the headphones with the Harman curve sound harsh to my ears around 5kHz, just as test results in 5128.

 

[FTB]

What headphones did you try ?

 

[DualTriode]

That aside, I don't know why averaging is promoted so much in these discussions. An average is a low pass filter of the dataset. By definition then, it reduces resolution. The reason it is often used is to make the data easier for humans to digest. But as targets, it just reduces specificity as we have seen in this paper.

This math and science kind of stuff especially when we are speaking of human population Psychophysics is the language of preference curves .

An average is required to let the exceptional shine.

Every step in the process includes averaging. Without a sample mean there would not be a correlation, a least square curve, a standard deviation, a standard error, split-plot ANOVA or a 2018 Harman Curve.

 

[Sha1rholder]

Anyone knows where to read the paper? I searched in AES, found nothing but an AES show.

 

[Dazerdoreal]

Interestingly, the headphones with the Harman curve sound harsh to my ears around 5kHz, just as test results in 5128.

Yeah this is not unusual at all. Measurements are not reliable for every ear. Some of my headphones also have peaks which I have not seen on any graph.

 

[FTB]

Yeah this is not unusual at all. Measurements are not reliable for every ear. Some of my headphones also have peaks which I have not seen on any graph.

The thing is a lot of measurements are badly done with these squinglink and stuff
A lot of reviewers use different equipment, different and unusual target etc.

 

[Dazerdoreal]

The thing is a lot of measurements are badly done with these squinglink and stuff
A lot of reviewers use different equipment, different and unusual target etc.

No, it is not just bad measurements. Maybe I just have an unusual HRTF.

 

[amirm]

Anyone knows where to read the paper? I searched in AES, found nothing but an AES show.

They have completely broken the search for recent papers at AES. My own paper doesn't come up unless you know how to search for it. It was a struggle finding the paper originally and I forgot how I did it. After 10 minutes of trying different things, I found it: https://aes2.org/publications/elibrary-page/?id=23068