Challenges to meaningful headphone measurement

[maxxevv]

What does the "XX dB Center " refer to there in the chart ?

 

[solderdude]

How it is perceived at 72dB SPL average. This is a comfortably loud level at which one can listen to a few songs before getting the urge to dial back again.
The other ones are how it is percieved at higher and lower levels so at 84dB SPL average (first blue line above the black one) it sounds pretty balanced. This is not by accident. It is very close at studio monitoring levels (between 82 and 85dB).
So HD800 should be listened to at reference levels.
Hoping one has hearing damage in the 6kHz area as the treble is piercing (to me) at these levels and need EQ badly.
At 60dB average levels (quiet long term home listening) it sounds bass-shy (first trace below the black)

When one compares the RAA plot to mine

It is easy to see how my plots indicate the tonal balance (no smoothing in my plot and heavy smoothing in RAA) of around 65dB average which is between the black trace and the one below.

Normally people don't listen to 40dB average or 100dB average so why they have so may traces is confusing. They are also incorrect in that perception correction is not linear. At higher levels it 'flattens'.
The RAA plot merely 'indicates' a bit what is happening but is technically an incorrect plot.

I like to see correlation for music listeners which are cautious with their hearing and want to know where actual peaks in the FR are.
That's the focus of what I try to do. Not just technical measurements.

 

[maxxevv]

I see. That "XX dB Center" is the SPL levels above and below the 72 dB reference level. Thanks for the kind explanation.
It certainly correlates quite well with my experience on them.

As for the 6K peak part, I find it generally alright for most of my listening but on the occasion, certain tracks will induce some discomfort in that treble region. Not often though and I'm moderately treble sensitive when I listen to IEMs. Peaky sounding IEMs are a no-no for me when it comes to long listening sessions.

 

[solderdude]

peaks in the treble area set of tinnitus in my case. That's probably why I am bound to find them and correlate it. HATS measurements have a tendency to obscure them, certainly when smoothed.
My initial goal for measuring started out for my own benefit but some folks liked to see them as well so I started a place where they could be posted not in a forum form.
I find it difficult to trace info on forums. When a thread becomes inactive the posted info basically becomes forgotten and hard to find.

Also I do not measure, just to measure but to correlate and above all to improve the performance in an as objectively possible way.

 

[Mad_Economist]

I meant the very sharp peaks around 10k as shown in post above (#15 from madeconomist). This is not corrected in any compensation and not in OW while at the same time it is caused by the Pinna but not corrected afterwards.

On my flatbed measurement rig I measure (and can hear) peaks in that frequency range or there are dips. These can be partly 'compensated' by a resonance in the test rig and thus not show the proper level.
Also the test rig's resonance frequencies may not be at the same frequencies as those of real humans or not be nearly as sharp.

When a headphone dips there it isn't as sound degrading as when there are peaks.
I am often told ... look you say you need to filter this out but I don't see it in this or that measurement. That is because they are masked by resonances in the test rig.

A number of fairly high-Q HRTF features exist in that band - with a sufficiently granular/unsmoothed HRTF, they may be compensated for well. Hammershøi & Møller discussed methodology for high resolution diffuse field measurements in one of their papers in the 90s (pardon me for a lack of link - it is late here presently), and of course free field data can be essentially arbitrarily fine.

For very short wavelengths, the presence of the pinna in the earpad is quite significant for the sound reaching the ear canal - at lower frequencies, it may be less significant, but it's much harder to argue this for the frequency range you're talking about, and the comparisons of earless designs to those with pinnae seem to affirm this.

As an aside re: equal loudness, it bears recalling that AFAIK the listening conditions (anechoic free field vs. headphone playback vs. diffuse field etc) impact the contours to some degree - if wanting to apply such a compensation to measurements, it might be pragmatic to dig into the lit on that front (full disclosure: not a topic I've ever looked into at meaningful length) rather than just taking the ISO 226 curves.

 

[JohnYang1997]

In general it's not only the person who performs the measurements that need to be extremely experienced in audio and knowing limitations and workarounds but also the people who read the measurements that need to be extremely skillful and unbiased while interpreting. In headphones measurements it's way serious than electronics, even though it's already really the case with electronics just even more so with headphones.

The thing is in general most knowledgeable people are not doing audio at least not at the consumer level. And headphones not only needs general audio knowledge but also actual listening and physics, acoustics so that's cross field sophisticated knowledge. More importantly the headphones being interacting to human directly it needs human for certain research. Not many people have the luck to do that. It's very difficult to meet all these the same time.

PS: people are much more subjective in headphones than electronics. Pushing objectivity into headphones aggressively is not really a good idea too. People can get extremely contradictory preferences. It's not like people can't hear difference between electronics in blind tests.

 

[solderdude]

PS: people are much more subjective in headphones than electronics. Pushing objectivity into headphones aggressively is not really a good idea too. People can get extremely contradictory preferences. It's not like people can't hear difference between electronics in blind tests.

And note that when differences are heard in well executed blind tests the differences are quite measurable already. Sometimes reaching 1 or 2 dB.
With headphones we are talking differences of several to tens of dB's in frequency response alone already.
Correlating the measurements needs shitloads of experience with many, many headphones. Acces to and experience with is paramount.
Misinterpreting 10dB sharp dips or peaks which may be measurement errors can easily happen when one misses the experience with lots of headphones and can't identify them as such. Training one self with equalizers and shoving up and down frequency bands with familar music and a good reference headphone to 'reset' your brain are essential tools to educate yourself.

 

[JohnYang1997]

And note that when differences are heard in well executed blind tests the differences are quite measurable already. Sometimes reaching 1 or 2 dB.
With headphones we are talking differences of several to tens of dB's in frequency response alone already.
Correlating the measurements needs shitloads of experience with many, many headphones. Acces to and experience with is paramount.
Misinterpreting 10dB sharp dips or peaks which may be measurement errors can easily happen when one misses the experience with lots of headphones and can't identify them as such. Training one self with equalizers and shoving up and down frequency bands with familar music and a good reference headphone to 'reset' your brain are essential tools to educate yourself.

Yes what you said were true.
But the point I was making is that the objectivity doesn't prove the subjective preference. Harman did some research on this. But to me the other candidates are too bad on their own and the result is not very overwhelming. However I'm not saying that there isn't an objective goal but the idea of aggressively pushing objective measurements without a balance in knowledge and information in other areas. When people can't relate what they interpret from measurements to what they hear they will call bullshit. So that's not very helpful.

Another point is. I did some non academic study with a few of my friends before. We are trained in such way we can always remember how a few tracks should sound we can eq any given headphones to 2db max deviation in the mid range without reference (reference can help just saying the way we were trained). If people are trained this way. I believe the preference of us will be close to 100%. Not mentioning headphone eq is not simple directly from measurements. All these are just my understanding of the current state of objective research is not mature nor ideal. We still have a long way to go. Also supporting last point.

 

[solderdude]

Simply having a headphone on your head in a different position compared to how it was measured can lead to audible differences.

Indeed training/experience helps as well.

Preference usually creates a substantial difference between how some people 'hear' a headphone compared to others and compared to measurements. I often read that folks find my measurements do not look remotely how they hear it and also there are those that find correlation very high.
This may be preference but may also be seal or the used kind of music or SPL differences.

So I agree correlation between any measurement and hearing may be ranging from high to virtually none.
That shouldn't defer anyone from trying though.

 

[pkane]

@solderdude or anyone who can answer, the target response curve from Harman is very up and down.
View attachment 28456

This is done measured in ear canals of listeners. The procedure was to measure in the ear canals of listeners the response when they listened to an EQ'd to flat speaker. Anyone know which speaker they used? Then they further let trained listeners listen to speakers and phones and make adjustments until the phones matched the sound of the speakers. Or I presume the mono speaker.

So okay if the measurements in the ear canal were done in the proper place while playing speakers that should work. Now you only need a measurement head unit with measurements that are equivalent. I don't know I'd think at some point they did measurements in the dummy head or measuring apparatus playing the same speakers. To close the circle if you will to show dummy head measures equal real human listener results. I'm guessing if I get the right paper maybe they show how much that varied between listeners who had the measuring mics in their ears?

Pretty easy to believe with the way phones are all over the place just this one step even if rather coarse could get us a target worth shooting for which is a big improvement over past methods. But with such a peak and fairly steep roll off I can imagine if the exact location of the peak moves very much at all between listeners you'd get some large subjective perceptual shift. So how close to the individual listener is this in practice, and how important is it?

I agree with SIY, that Sonarworks correction seems to do a great job. They don't measure this at all like anyone else. You can puzzle thru the patent here for some idea.
https://patents.google.com/patent/US10021484B2/en

Seems to me Harman would be well advised to use this measurement method and then correlate results to their reference speaker. Maybe they've got other knowledge and find it not needed.

Thinking about all this points out a way I could develop my own EQ curve with my own phones and ears. But it would be tedious as hell. I could play pink noise over my preferred speakers, and EQ so the phones sound the same. You'd likely need to filter the noise into bands to get it right. Probably break it into ERB bands or 1/3 octaves at least. Matching actual perceived sound levels each step of course. That would take some doing. If I liked listening to headphones I'd do it.

I attempted to measure my HRTF response using in-ear mics. It would seem that for me there's a difference between left and right, and the peak is at 4kHz compared to Harman's 3...

 

[solderdude]

HRTF is not only Pinna effects but includes the ear canal as well which is blocked by the in ear mic which also isn't flush with the ear canal entrance so you may only have a partial HRTF which doesn't even remotely look like that of raw HATS measurements.
And it says little to nothing about how the inner ear works either.

EQ-ing towards it may sound 'better' or may not. Depends on multiple factors.

Did you try Griesingers method as well and compared results ?
That would make a compelling case when it did.
Using 2 different methods to arrive at the same EQ/conclusion.


I think that HRTF isn't as important as some make it out to be because the brain takes HRTF into account (calibrates it out) and it also differs from determining HRTF from speakers as sounds come from the side.

The actual 'acoustical' impedance also will differ from a HATS.

Personally I don't worry about HRTF and think it is inconsequential and one certainly should not EQ to it as the brain already EQ's it out.
I am probably wrong about this but that's what it looks like from the angle I see (everything is perception based on personal experience and views one has)