Headphone testing: To Pinna or Not to Pinna (that is the question)

[xnor]

I'm not a fan of these headphone CSDs either. Just wanted to comment on it.
 
I think that it's not comparable to measurements done under anechoic conditions. To make them more comparable the non-pinna measurement is probably the better choice and the simple plate, as mentioned in post #1, should have some absorbing material.
 
Yeah, the pinna and ear-canal could have some resonances that show up in the CSD. I also think that these resonant frequencies (they call it "ridges in the CSD") change depending on the headphone construction.

 

[Nevod]

Quote:

Nevod, would I get in trouble if I send you one or two of the AES papers?  I hate to see you not able to read them!  OR, you can buy them pretty cheap on the AES site.

You certainly wouldn't be got in trouble by me personally, but I don't know how well AES chases people who distribute their docs. 20$ isn't so cheap for me, moreso that one paper won't be enough, there seems to be quite many of them related to the topic. I though do have plans on once scraping together some money and getting the papers.
 
As for the leaks, I too think they aren't worth a hassle, after all, they're a known and well understandable stuff.
As for the measurements.. Well, that depends mainly on what do you want, IMO. If you just want a comparison of headphones done on your equipment, then indeed, pinna & DRP is the way to go, but still you'd have to calibrate it against something you use as etalon. That may be a flat-FR speaker 30 degrees off-axis at 3 meter distance in a anechoic room for free-field-like etalon, may be the same in an diffusely reverberant room for diffuse-field-like, may be a flat-FR speaker right on shoulder for binaural-like etalon.. I'll add some more a bit later, have to run.
 
xnor,
I think that it would be actually good to do both CSD and FR measurements with two kinds of baffles - reflectant and absorbant - the difference beetwen them would show how well a headphone's ear chamber absorbs reflections. 

 

[Nevod]

Took me a long time to get my thoughts together.
 
First, on the issue of PDR and corresponding responce shift: It is beneficial to have PDR or acoustic impedance plot for a headphone anyway, as it helps general public and particularly manufacturer assess some deficiencies in design which may cause problematic responce due to PDR and not due to driver's FR itself. Also, having that plot would allow people to guess if the headphone could cause personal problems, due to interaction of PDR with their own ear canals's resonances. Of course, without having the plot of own's ear canal FR, it's not possible to say for sure wether that headphone would be suitable for the person or not, but it's possible to tell that a HP with heavily jagged PDR could cause problems if ear canal would overlap with PDR resonance, causing a resonance that might be dangerous.
 
Other thought is more aimed at the AES paper of the next thread - while it finds a more pleasurable responce, is that responce really more correct? The tests to find the correctness of the responce for a given case (binaural, diffuse field emulation, free field emulation) should ba based on correctness of determining pitch and location of a reproduced sound source, not just preference. Though, without having seen the paper, I can't object really.

 

[Tonmeister2008]

I'm not sure I'd agree that PDR is not a problem with circumaural headphones.  True, the data shown in the original Muller/Hammershoi papers does seem to show modest variations, though in the "Transfer Characteristsics..." paper they only show up to 7 kHz (and PDF is increasing with freq. up to that point).  Also of course it depends on the headphones.  Dorte Hammershoi was one of my profs at Aalborg, and she has repeatedly reminded me that just because a headphone is "open" does not make it FEC, and vise versa.  Hard to know which headphones will cause problems.

Florian Volk at TU Munich has also done some measurements of PDR and is measuring a bunch more headphones as part of a project.  it will be interesting to see what he comes up with.

Nevod, would I get in trouble if I send you one or two of the AES papers?  I hate to see you not able to read them!  OR, you can buy them pretty cheap on the AES site.

You might get in trouble for breaking AES copyright. As AES Prsident Elect I might be obligated to have your AES membership revoked if you give out free copies of AES papers

 

[ComfyGrados]

Quote:

the pinnae you will use are "average" pinnae, so not correct for any one person.

 
I believe it may be possible, and very simply so, to come up with a dummy head measuring technique which will result in measurements which will be accurate for people with all sizes and shapes of heads and ears:
 
1. Measure a confirmed flat speaker system in an ideal listening environment with your dummy head.
2. Use the resulting frequency response graph as your HRTF compensation. The resulting frequency response is how your dummy ears perceive neutral audio. Headphones which your dummy ears perceive to sound identical to the speaker system, then, should be neutral as well.
3. Measure headphones with dummy head, then add your HRTF compensation which you came up with in step 2.
 
Thoughts?

 

[ultrabike]

Quote:

 
I believe it may be possible, and very simply so, to come up with a dummy head measuring technique which will result in measurements which will be accurate for people with all sizes and shapes of heads and ears:
 
1. Measure a confirmed flat speaker system in an ideal listening environment with your dummy head.
2. Use the resulting frequency response graph as your HRTF compensation. The resulting frequency response is how your dummy ears perceive neutral audio. Headphones which your dummy ears perceive to sound identical to the speaker system, then, should be neutral as well.
3. Measure headphones with dummy head, then add your HRTF compensation which you came up with in step 2.
 
Thoughts?

 
Sounds reasonable to me, but I'm no expert.

 

[MeatusMaximus]

Quote:

 
I believe it may be possible, and very simply so, to come up with a dummy head measuring technique which will result in measurements which will be accurate for people with all sizes and shapes of heads and ears:
 
1. Measure a confirmed flat speaker system in an ideal listening environment with your dummy head.
2. Use the resulting frequency response graph as your HRTF compensation. The resulting frequency response is how your dummy ears perceive neutral audio. Headphones which your dummy ears perceive to sound identical to the speaker system, then, should be neutral as well.
3. Measure headphones with dummy head, then add your HRTF compensation which you came up with in step 2.
 
Thoughts?

This is essentially what we are doing (see "The Relationship between Perception and Measurement of Headphone Sound Quality"  Sean Olive and Todd Welti,, AES preprint 8744, AES 133rd convention.  However, I'm not sure I see why that is accurate for people with all sizes and shapes of heads and ears.  unless when you say "dummy head" you mean our own heads (insert joke here).

 

[Nevod]

Quote:

Thoughts?

If I undestand correctly, headphone measurements are already compensated that way, mapped to diffuse-field responce (usually) or free-field responce(rarely). One who performed such compensation curve creation can also measure that test headphone on a flat baffle and get a transfer function which would allow other people using similarly set-up baffles to see equalised responce.
There's however a problem that there are some ear canal resonance effects which are triggered just by presence of headphones by head, they differ by person and by headphone's construction, so one wouldn't be able to see how _exactly_ headphone would produce sound for a person. A headphone that is flat by above means could have some mad spikes in FR in 6-11 kHz range.
 
I personally strive for a headphone with a straight-flat uncompensated responce, that is optimal for binaural reproduction, although pretty much requires use of a crossfeed at least.

 

[ComfyGrados]

Quote:

However, I'm not sure I see why that is accurate for people with all sizes and shapes of heads and ears.

 
I'm thinking headphone measurements using such an HRTF compensation method would be accurate relative to the flat speaker reference rather than accurate relative to peoples' sizes and shapes of heads and ears.
 
If you have a headphone which a dummy head measures as sounding identical to a flat speaker reference, then would not that headphone sound exactly like (in terms of frequency response) the flat speaker reference to people of all shapes and sizes of heads, not accounting for the "missing 6 decibels" effect? If not... why? If not... you say this is essentially what you're doing anyways, so if not, do you think that's "as accurate as we're gonna get?"
 
I'm thinking that while using an HRTF compensation method such as an average human ear diffuse field EQ curve in order to produce HRTF-compensated headphone measurements would result in measurements which are accurate relative only to a mean average of human ears, using an HRTF compensation based on the dummy head's specific HRTF relative to a flat speaker source would result in measurements which are accurate relative to a fixed, definable point, e.g. the flat speakers.
 
Do you understand what I'm trying to say?
 
 
Quote:

If I undestand correctly, headphone measurements are already compensated that way

 
And if I understand correctly, not all headphone manufacturers compensate their measurements that way... don't some headphone manufacturers aim for, for example, a target diffuse field curve like a mean average human ear diffuse field curve?
 
After all, the most important ramifications of this sort of discussion and research, in terms of what our ears think, is it all some day leading to better sounding headphones!

 

[ultrabike]

If I undestand correctly, headphone measurements are already compensated that way, mapped to diffuse-field responce (usually) or free-field responce(rarely). One who performed such compensation curve creation can also measure that test headphone on a flat baffle and get a transfer function which would allow other people using similarly set-up baffles to see equalised responce.
There's however a problem that there are some ear canal resonance effects which are triggered just by presence of headphones by head, they differ by person and by headphone's construction, so one wouldn't be able to see how _exactly_ headphone would produce sound for a person. A headphone that is flat by above means could have some mad spikes in FR in 6-11 kHz range.
 
I personally strive for a headphone with a straight-flat uncompensated responce, that is optimal for binaural reproduction, although pretty much requires use of a crossfeed at least.
 

Assume a headphone is designed to reproduce the same frequency response that a flat speaker produces when measured using mics in dummy head 1. 
 
If I understand things correctly, if this headphone is placed on a different dummy head 2, the headphone's measured frequency response will be different from the response obtained through using the same flat measuring speakers using mics in dummy head 2.
 
If so, here are my noob questions:
 
1) Is the cause of these differences a result of the variation in "acoustic impedance" when using headphone on head 1 and 2?
2) Is "acoustic impedance" variation a result of the enclosure volume difference (and air pressure) between head 1 + headphone and head 2 + headphone?
3) If this is so, what is the enclosure volume variance and distribution across individuals?
4) What is the variance vs frequency, of the frequency response deviation from flat when using an arbitrary head 1 "flat" measuring headphone on a representative set of distinct heads?
5) How does these differences correlate with sound quality perception?
 
Are all these questions addressed concisely by the paper or/and some other papers? ... I just want to stop being a noob, or at least diminish my noobness level.
 
BTW, I also seem to like flat uncompensated frequency response (or at least not too much compensation), and I can see how binaural recordings would benefit from this, even if things are not perfect 

 
I also wonder how close the Smyth Realiser is in achieving optimal results with regular and multi-track recordings. Seems they standardized their product to work with the SRS-2170, and may offer some HRTF customization.

 

[Nevod]

And if I understand correctly, not all headphone manufacturers compensate their measurements that way... don't some headphone manufacturers aim for, for example, a target diffuse field curve like a mean average human ear diffuse field curve?
 
After all, the most important ramifications of this sort of discussion and research, in terms of what our ears think, is it all some day leading to better sounding headphones!

Just try to draw a couple graphs and you'll see it's not universally applicable. If a headphone has responce that sound identical to a flat speaker for a head, that only means its transfer function is identical to head's HRTF.
See:
Let's assume that a DF frequency responce for head 1 has a slight hill at 5kHz (measured by your 1) ). Let's assume we have a headphone which matches that FR exactly. Subtracting the 1) FR from headphone's FR you'll get a perfect straight line. Looks perfect, right?
Now take a head 2.. And by some virtue that one has 2 sharp peaks at 2 and 3 kHz, just for example. Now try to subtract that from the headphone's uncompensated FR: you get 2 pits at 2 and 3 kHz and a slight hill at 5 kHz. 
 
So it's more correct to have a personal HRTF. Would be neat if hardware stores would provide you with an HRTF measurement if you buy headphones from them.. Though it's not impossible at home either.
 

Quote:

 
Assume a headphone is designed to reproduce the same frequency response that a flat speaker produces when measured using mics in dummy head 1. 
 
If I understand things correctly, if this headphone is placed on a different dummy head 2, the headphone's measured frequency response will be different from the response obtained through using the same flat measuring speakers using mics in dummy head 2.
 
If so, here are my noob questions:
 
1) Is the cause of these differences a result of the variation in "acoustic impedance" when using headphone on head 1 and 2?
2) Is "acoustic impedance" variation a result of the enclosure volume difference (and air pressure) between head 1 + headphone and head 2 + headphone?
3) If this is so, what is the enclosure volume variance and distribution across individuals?
4) What is the variance vs frequency, of the frequency response deviation from flat when using an arbitrary head 1 "flat" measuring headphone on a representative set of distinct heads?
5) How does these differences correlate with sound quality perception?
 
Are all these questions addressed concisely by the paper or/and some other papers? ... I just want to stop being a noob, or at least diminish my noobness level.
 
BTW, I also seem to like flat uncompensated frequency response (or at least not too much compensation), and I can see how binaural recordings would benefit from this, even if things are not perfect 

 
I also wonder how close the Smyth Realiser is in achieving optimal results with regular and multi-track recordings. Seems they standardized their product to work with the SRS-2170, and may offer some HRTF customization.

I must warn that I'm no specialist either! I first have to at least thoroughly study papers MM sent to me, not just read them quickly. So what I'll say next is a supposition, not truth.
 
First, it's even more variable than that. Even if two headphones measure exactly the same on one head (FR-PR-square-impulse etc), they could measure differently from each other on another head.
1) Seems that it is indeed caused by variation of acoustic impedance, which causes different resonances to arise. Possible resonances are determined by exact pinna and canal structure. Equal impedance + equal IR seems to result in truly equal sound.
2) Enclosure volume plays a role to, but mostly it seems proportional to how much of the open space around is shrouded by headphone's structure.
3) Don't know. Variation in personal ear resonances seem to be very high from what I've read, though sources aren't very reliable.
4) In Moller's paper, all tested headphones had under +-6dB difference under 6kHz, and measurement of variation over 6 kHz was deemed impractical (I have to reread on that! maybe I misunderstood that part)
5) Seems to be more important to localisation than tonal balance perception, but high peaks would probably cause brightness of sound ( so a HP bright for one may be alright for another one even objectively)
 
These questions are addressed in papers, but I won't say if the extent of how well they're addressed is good enough for me. 
 
Staxes seem to have one of the lowest acoustic impedances, as they're literally very acoustically transparent, so they're well suited for that task.

 

[MeatusMaximus]

Quote:

 
 
If you have a headphone which a dummy head measures as sounding identical to a flat speaker reference, then would not that headphone sound exactly like (in terms of frequency response) the flat speaker reference to people of all shapes and sizes of heads, not accounting for the "missing 6 decibels" effect? If not... why? If not... you say this is essentially what you're doing anyways, so if not, do you think that's "as accurate as we're gonna get?"
 
 

Yes, but different people would have different responses in the diffuse field, and yet different responses with headphones.  For any given person, if this difference was the same for the DF condition  and the headphone condition, then it would work as you say.  BUT, i don't assume this is the case.  The interaction of the headphone and pinna/canal could very well be different for DF and headphone conditions.

 

[MeatusMaximus]

Quote:

 
If so, here are my noob questions:
 
1) Is the cause of these differences a result of the variation in "acoustic impedance" when using headphone on head 1 and 2?
2) Is "acoustic impedance" variation a result of the enclosure volume difference (and air pressure) between head 1 + headphone and head 2 + headphone?
3) If this is so, what is the enclosure volume variance and distribution across individuals?
4) What is the variance vs frequency, of the frequency response deviation from flat when using an arbitrary head 1 "flat" measuring headphone on a representative set of distinct heads?
5) How does these differences correlate with sound quality perception?
 
 

My understanding: Acoustic impedance does have effect, small to moderate for some phones, increasing at higher frequencies, but also increasingly difficult to measure.  LEAKAGE is the elephant in the room.  We have seen HUGE effects for leakage, particularly closed supraaural phones, but even circumaural.  It varies from person to person and even placement to placement.  THat affects <1 kHz mostly.  How they correlate is the big question.  I believe they have an effect all right, but hard to quantify.

 

[ultrabike]

Acoustic impedance and leakage aside, what other factors might contribute to differences in headphone sound quality perception?
 
I think I'm beginning to "digest" some of these concepts better. I understood what ComfyGrados was saying but after some more thought... consider two distinct dummy heads (say 1 and 2) with different HRTFs. Say headphone 1 is optimized for dummy head 1. When the dummy head 1 optimized headphone is applied to dummy head 2, it effectively bypasses dummy head 2 HRTF and applies dummy head 1 HRTF. This would lead to a difference in perception of neutral from dummy head 2 point of view. I guess how much of the HRTF is bypassed depends on the type and construction of the headphone.
 
Is the above correct? If so, are these differences in HRTF less dominant when compared to differences in acoustic impedance and leakage? How do variances in HRTF affect circumaural vs IEMs? Thanks

 

[MeatusMaximus]

Quote:

If so, are these differences in HRTF less dominant when compared to differences in acoustic impedance and leakage?

I think this is closely related to the question i started this thread with.  Is the error of using a standard pinna/ear canal greater than the error of not including them.  I don't know for sure.