Master Complaint Thread About Headphone Measurements

[Heinrich]

Just searched about the diffuse field and found this: (declaration that I am no expert nor with a ton of knowledge about the plots)
View attachment 99576
https://forum.hifiguides.com/t/grap...n-ear-over-ear-and-harman-linear-in-room/9270

it seems the Harman Linear in room is a response measured at ear drum level with dummy head mics in a treated room and anechoic chamber neutral speaker, while the green and orange are their target curve for in ear and over ear phones.

to me the big players for neutralish sounding headphones are aiming at the linear in room curve which is like 8-12db lower than the target curve bass and with a 4db boost around 3khz, so shall we compare the measured headphone curve to that and make some comments on subject listening section to both tunings?

This is a really strange topic, and I like it. Where else in audio do we have the freedom left to talk about subjective feelings?

I've read the original article by (sorry for all the umlauts, only available in German, me thinks) Günther Theile, "Beurteilungskriterien für Kopfhörer unter Berücksichtigung ..." (https://hauptmikrofon.de/theile/1985-6_Kopfhörer-Beurteilungskriterien_NTG-1985.pdf), 1985

His argument goes like this, addressing stereo:
- the human hearing cannot use its inherent urge to actively seek for the acoustic source, e/g head movement will give irritating cues
- the speakers should be devoid of any coloration that may be missinterpreted as an effect of inclination angle
- freefield compensation comprises directional cues, in that the hearing identifies "easily" (GT) the spectral signature of sound arriving from the front**

Conclusion: freefield compensation doesn't hold, hence (without any further argumentation I could identify) diffuse field** is it

I personally would agree with the first. Second and third assumption are understandable from a scientist's perspective. Once I know the spectral distribution of an acoustic event, I could match that pattern to a second event and may conclude on a change in inclination angle. In physiocological tests repeated, hence same signals are used which makes the point.

But that isn't true when listening to real music. Music is about the unknown, and even when known the position is kept fixed (except for those infamous dedicated effect recordings).

Btw: the black line in the above graphic shows a freefield compensation. All Harman's are diffuse field, basically. The green line for in-ear may compensate @8kHz, 3dB for the missing outer ear aka 'pinna' interaction, that the over-ear, orange, presumably provides.

I think the theory is a bit flawed right from the start, even if the results are admittedly good enough after possible 'corrections'. We could and should appreciate the shift to a preference based perspective ;-)


** freefield compensation: a single speaker is set in front of a listener (or a dummy) and a match between speaker free air spl versus in ear spl is done, subjectively or by measurement, which relation is taken as a target when measuring headphones on a dummy head/ear; diffuse field the same but with speakers, devoid of Harman tilt me thinks, all around the head

 

[isostasy]

I have a potentially inane question...

Has anyone else noticed that the "Harman Filters" available in some squig link sites don't match the OE 2015 target? Here's what I mean:

This is what you get when you set "Harman Filters" to "Harman 2018 Filters" then overlay "Harman AE/OE 2018". They align pretty much perfectly. This is also the case with "Harman 2013 Filters".

But with "Harman 2015 Filters" they're different:

It's only about 1dB and seems inconsequential but the reason I'm interested is these filters are used on sites with 5128 measurements too. And what do you get when you apply what is called "Harman 2015 Filters"?

Almost exactly "Harman Beta (2024)". Except those filters are applied to 5128 DF, not a 5128 equivalent of "Harman in-room", because they haven't done that yet afaik. So you if you back and apply these filters to the DF response of the GRAS 43AG-7 with KB-50XX pinna you get:

...which as you can see is quite a way off the OE 2018 target. I find this interesting because if you then use comparable filters to compare headphone measurements, for example something from the Sennheiser HD6xx series, you'll find results are quite different:

vs.

Whilst I'm on the subject, does anybody have any idea what the "High Res Harman 2018" target really is? I originally saw it in a Headphones.com YouTube video are article from what I remember. At the time I think they said it was an "unsmoothed" version of the OE 2018 target. Yet this clearly isn't the case:

8kHz+ the difference could be smoothing but the "ear gain" peak at 3kHz is clearly higher.

Again if you say I'm overthinking it I agree, but I find it interesting nonetheless.

 

[Heinrich]

Again if you say I'm overthinking it I agree, but I find it interesting nonetheless.

The industry did a very good job in confusing people. One has to read all technical papers on the „compensation targets“. Avoid any second hand explanation. Otherwise their use is erratic. This is not just a rant.

We do *not* need another „Harman“ but an utterly unsmoothed, detailed amplitude frequency response curve for the HATS (head and torso simulator) used for a measurement, taken in real, physical diffuse field and free field. That *is* the reference one might compare a headphone measurement to, not a „Harman“ or other „target“. That is the first incremental step to match a headphone to a human head. Second one might argue, if, say, the torso of each and another person differ that much, acoustically. Again, second incremental approach after the first and ongoing.

I‘m under the layman‘s impression that the industry forgot about fundamental best practices—when trying to solve the problem as a whole by averaging over the variety of personal features of people … or smoothing out even the HATS‘s specifics, then saying: „ignore“?!

 

[Curupira]

The industry did a very good job in confusing people. One has to read all technical papers on the „compensation targets“. Avoid any second hand explanation. Otherwise their use is erratic. This is not just a rant.

We do *not* need another „Harman“ but an utterly unsmoothed, detailed amplitude frequency response curve for the HATS (head and torso simulator) used for a measurement, taken in real, physical diffuse field and free field. That *is* the reference one might compare a headphone measurement to, not a „Harman“ or other „target“. That is the first incremental step to match a headphone to a human head. Second one might argue, if, say, the torso of each and another person differ that much, acoustically. Again, second incremental approach after the first and ongoing.

I‘m under the layman‘s impression that the industry forgot about fundamental best practices—when trying to solve the problem as a whole by averaging over the variety of personal features of people … or smoothing out even the HATS‘s specifics, then saying: „ignore“?!

Wasn't this the state of headphone measurements from the 1980's right up to Olive et al's research ? Wasn't Harman such a magic bullet inside academia BECAUSE they dismantled the old paradigm of free field and diffuse field anechoic measurements as an ending point of reference for HRTF?

 

[markanini]

Wasn't this the state of headphone measurements from the 1980's right up to Olive et al's research ? Wasn't Harman such a magic bullet inside academia BECAUSE they dismantled the old paradigm of free field and diffuse field anechoic measurements as an ending point of reference for HRTF?

Measuring a representative listening room as a base and validating preference adjustments is a great leap forward. On the flipside the overly smoothed curve would be considered dated today and has fed into misconceptions about what good target compliance should be. Emerging updates to measurements standards calls for Sean Olives work to be redone and we don't know yet if he or someone else will take that upon themselves. Rig-specific unsmoothed DF curves can be useful in the interim although they lack validation with preference adjustments applied that would make them equivalent to Harman 2018.

 

[Heinrich]

Wasn't this the state of headphone measurements from the 1980's right up to Olive et al's research ? Wasn't Harman such a magic bullet inside academia BECAUSE they dismantled the old paradigm of free field and diffuse field anechoic measurements as an ending point of reference for HRTF?

Not quite, as I exemplified here before, see my take on Günther Theile. Point in case is, as far as I‘m concerned, that just semi costly investigations are missing, literature is, may I say ‚hidden‘ behind paywalls. The public is held uninformed, and Harman seems to play, in bigger parts rightfully so, the ‚preference‘ card. I might even refer to Harman as a player who gives talks on *not* investigating.

I‘m not after the general shape of the HRTF, but after individual details. Even an HATS has a head and torso relatedvtransfer function when exposed to whatever external sound field. The headphone has to replicate the particular HRTF when seeted on that HATS. If seeted on another HATS, you guess already, that other HRTF— in all the nasty details, has to be replicated. By the very same headphone. Look out for literature on especially that, and find nothing.

Reiterated, presumably the task is achievable, at least incrementally. That would be a totally different approach than—averaging over and smoothing away details, leaving the rest deliberately to literally, ignorance.

 

[ilomaenkimi]

When testing headphones here, listening test should do first. Measurements after listening.

 

[oleg87]

When testing headphones here, listening test should do first. Measurements after listening.

...why? I can read people's subjective impressions literally anywhere on the internet. The value is the measurements taken with equipment way outside my budget.

 

[ilomaenkimi]

...why? I can read people's subjective impressions literally anywhere on the internet. The value is the measurements taken with equipment way outside my budget.

Seeing measurements before listening affects listening impressions.
Or is it because there's possibility to make fool of yourself?

 

[amirm]

Seeing measurements before listening affects listening impressions.

And seeing the headphone, knowing its brand/cost, impression of reading other people's opinions, doesn't do that???

If you want something to bias you, it might as well be something reliable like measurements.

 

[ilomaenkimi]

And seeing the headphone, knowing its brand/cost, impression of reading other people's opinions, doesn't do that???

Not as much seeing measurements before listening.
You see massive 5khz bumb, what draws you attention at listening?
Or does it creates shame if you listen before measurements and that bumb remains unnoticed?
Reputation..

 

[Jimbob54]

Not as much seeing measurements before listening.
You see massive 5khz bumb, what draws you attention at listening?
Or does it creates shame if you listen before measurements and that bumb remains unnoticed?
Reputation..

You realise any reviewer publishes and listening regardless of the order ?

 

[Doodski]

Not as much seeing measurements before listening.
You see massive 5khz bumb, what draws you attention at listening?
Or does it creates shame if you listen before measurements and that bumb remains unnoticed?
Reputation..

No worries @amirm this person is causing issues in another thread too. Obviously here to dig and cause issues.

Reality check - is this YouTube Noire X review total BS or not?

I'm semi-interested in the DCA Noire X and have read/watched a bunch of reviews. I'm still scratching my head over this one: To cut to the chase, begin watching at 24:50 - I don't know what to make of his comments about the Noire X's lack of resolution and that it won't "scale" up as you...

www.audiosciencereview.com

 

[solderdude]

Whilst I'm on the subject, does anybody have any idea what the "High Res Harman 2018" target really is? I originally saw it in a Headphones.com YouTube video are article from what I remember. At the time I think they said it was an "unsmoothed" version of the OE 2018 target. Yet this clearly isn't the case:

It was an attempt to get a less 'smoothed' target closer to the actual average response on that specific fixture. As fixtures usually have a higher ear-gain than humans the target was higher. The dip between 8kHz and 10kHz also more visible. It is really there in fixtures (and human heads) so should be in the target.
The overly smoothed and probably obtained by averaging original curve clearly is too low. Just look at all the compensated plots from Amir for instance and it appears as though all headphones have a substantial upper treble extension so the curve is obviously incorrect there.

So this was an attempt to get a more accurate target curve.
The thing is the response above 8kHz or so is inaccurate anyway and thus is best left 'smoothed' on average but should be higher in level.
Even better.... the response below 50Hz and above 8kHz (or 10kHz for the newer fixtures) should best be greyed out with remarks of inaccuracy.

 

[amirm]

You see massive 5khz bumb, what draws you attention at listening?

How do you know what a 5 kHz bump sounds like? You have specifically been trained that way? If not, you will have no chance of identifying that deficiency in listening tests. I know, I have taken Harman training tests in front of top high-end dealers and watch them fail miserably in this regard.

What is super valuable about this information is when you apply it to equalization. You create an inverse filter for that bump, then turn it on and off, blind if necessary. Now you instantly learn a) what that bump sounds like and b) whether you like it corrected. And to what extent. This is why I measure before listening. It gives me this incredibly useful information to detect audibility of frequency response impairments.

Please watch this video for more detail:

 

[Heinrich]

It was an attempt to get a less 'smoothed' target closer to the actual average response on that specific fixture. As fixtures usually have a higher ear-gain than humans the target was higher. The dip between 8kHz and 10kHz also more visible. It is really there in fixtures (and human heads) so should be in the target.
The overly smoothed and probably obtained by averaging original curve clearly is too low. Just look at all the compensated plots from Amir for instance and it appears as though all headphones have a substantial upper treble extension so the curve is obviously incorrect there.

So this was an attempt to get a more accurate target curve.
The thing is the response above 8kHz or so is inaccurate anyway and thus is best left 'smoothed' on average but should be higher in level.
Even better.... the response below 50Hz and above 8kHz (or 10kHz for the newer fixtures) should best be greyed out with remarks of inaccuracy.

What I said. First step is to not smooth the HATS response (to free/diffuse/any field) at all and compare to the response to a headphone/IEM's output. If the HATS is not that typical, never mind. Second step is to evaluate the variability of responses to the same headphone on real peoples' heads. Third do some educated guesswork on how wide the bandwith of deviations is. Try to categorize deviations (e/g "on big ears so and so, on shallow ears so and so, deep ear canal so and so"). Please show me the results (this is a question to Harman) on my side of the usual paywalls

As a motivation, the less smoothed version of "The Harman" shows a 10k trough followed by a 14k peak. Those features are critizised a lot when IEMs realize them, bad cheap IEM! People eagerly equalize them out and prefer, go figure.

 

[isostasy]

It was an attempt to get a less 'smoothed' target closer to the actual average response on that specific fixture. As fixtures usually have a higher ear-gain than humans the target was higher. The dip between 8kHz and 10kHz also more visible. It is really there in fixtures (and human heads) so should be in the target.
The overly smoothed and probably obtained by averaging original curve clearly is too low. Just look at all the compensated plots from Amir for instance and it appears as though all headphones have a substantial upper treble extension so the curve is obviously incorrect there.

So this was an attempt to get a more accurate target curve.
The thing is the response above 8kHz or so is inaccurate anyway and thus is best left 'smoothed' on average but should be higher in level.
Even better.... the response below 50Hz and above 8kHz (or 10kHz for the newer fixtures) should best be greyed out with remarks of inaccuracy.

That sounds like a useful aim, and having that dip has certainly helped me EQ treble without being mislead. I still don't understand why the ear-gain should be so much higher though if the smoothed Harman curve was derived on the same fixture? I feel like I'm missing something

 

[solderdude]

These fixtures tend to exaggerate the ear gain compared to actual human ears.

 

[amirm]

First step is to not smooth the HATS response (to free/diffuse/any field) at all and compare to the response to a headphone/IEM's output.

I smooth my measurements less than research/target to see more detail. Going beyond that will give you perceptually false information because our hearing bandwidth enlarges as frequencies go up so a lot of variations are not audible. This is true for both speakers and headphones.

 

[amirm]

Second step is to evaluate the variability of responses to the same headphone on real peoples' heads. Third do some educated guesswork on how wide the bandwith of deviations is. Try to categorize deviations (e/g "on big ears so and so, on shallow ears so and so, deep ear canal so and so"). Please show me the results (this is a question to Harman) on my side of the usual paywalls

You are talking about a research project, not a review.