DIY Headphone Measurements - Ear Simulation?

[mnemonix]

Apologies for what are probably very basic questions. I assume we're looking at the measurement of a constant volume (spl?), frequency sweep here. What is actually producing the sound in the case above? does it matter or are we only interested in relative performance of the measuring devices? Since I assume both the sound producing device (headphone/speaker) AND the sound receiving device (mic) will have their own non linear frequency response, coupled with the variation produced by the physical anatomy of the measuring setup/real listeners heads, how are these kind of measurements actually useful to anyone since there appear to be too many variables? Assuming they are useful, how are they normalised so comparisons can be made?

If there's a text available that's understandable to a non-audio engineer can anyone point me to it?! Thanks!

 

[Deleted member 16543]

Apologies for what are probably very basic questions. I assume we're looking at the measurement of a constant volume (spl?), frequency sweep here.

Most likely the test signal is a constant amplitude sine sweep, yes.

What is actually producing the sound in the case above?

A combination of transducer and receiver electroacoustic characteristics in relation to their mutual location and their surrounding environment geometry, plus whatever the algorithm is doing to smooth out the raw response.

does it matter or are we only interested in relative performance of the measuring devices?

It depends on your goal. If you want to make or buy something (possibly cheaper) that gives you the same response as a widely recognized standard coupler then comparison between devices is enough.
However, that response will not directly translate into sonic information unless the surrounding geometry closely matches the one that is actually there during listening.

Since I assume both the sound producing device (headphone/speaker) AND the sound receiving device (mic) will have their own non linear frequency response, coupled with the variation produced by the physical anatomy of the measuring setup/real listeners heads, how are these kind of measurements actually useful to anyone since there appear to be too many variables? Assuming they are useful, how are they normalised so comparisons can be made?

That's why it's important to choose a standard measurement apparatus that uses a close replica of the head up to our own sensing element (the eardrum).
I really hope to see those anatomically accurate ears become the new standard. Of course, there's already 2 of them and I'm pretty sure none of those companies is likely to concede the other should be the de facto standard. They didn't spend time and money only for the glory of making a product that works but is not recognized as 'exactly' what the industry uses (unlike some guy I know ). Just like with binaural microphones with closed ear canals, it's unlikely we'll ever see a 'winner' in the anatomically accurate camp.
Provided there ever will be an agreement on which model of anatomically accurate ear is best (there won't be, ever), then the mic non flatness of the FR can be accounted for in the smoothing algorithm.
The smoothing algorithm is a big part of translating the raw response into information that is sonically relevant and accurate.
I like the software Acourate for exactly this very reason. It also fixes the capsule FR by letting you input its calibration file (just like many other softwares).
The only thing that remains to do now is decide what that target response should look like. Harman approach makes the most sense to me. They just didn't use the right ears (nor the right smoothing algorithm, most likely), but targeting what a balanced speaker measures like at the listening position is the best approach to me, conceptually speaking, since we more or less already know what a balanced speaker measures like at the listening position, as measured from a standard, non binaural measurement mic.

If there's a text available that's understandable to a non-audio engineer can anyone point me to it?! Thanks!

The average audio engineer knows nothing about this stuff as it's probably way above their head. Not because they're stupid, but because they don't really need to know. They are artistic guys, not scientific guys. Measurements are about science, not art.
Sure, you may find the scientific literate audio engineer here and there, but I wouldn't expect that to be the norm. It's like with painters. There is the once in a blue moon guy like Da Vinci, who not only paints artistically but is also a scientist, but most of them don't know the first thing about the physics of light absorption, the mechanisms that generate colors, and how our sight really works.
The book I linked to is simple enough in my opinion, but it does require some general understanding of Fourier transform, cross overs and audio related stuff. Nothing that should be too difficult for someone that isn't totally new to those matters, though. There's no advanced math in there, for one. It doesn't talk about headphone measurements, but the psycho acoustics principles at play are similar if not exactly the same. It's just a matter of replacing a standard measurement mic with a binaural one, and use speakers that have already been equalized to see what the binaural mic measures like.

 

[Dreyfus]

I guess my point is that whatever measurements you make out of a rig that doesn't closely match the ear (+head for speakers measurement) is going to have loose correlation to the actual perceived sound quality.

It all depends on your goals and ambitions at the end. If you want to measure what you (physically) hear, there is no other way than starting with an anatomically correct replication of your own ears, indeed. Since that is hard to accomplish for the various reasons we already gathered, the best we can get is a system whose measurements have no absolute weight but allow comparative evaluations at least. Then, the most important aspects for such a rig are probably repeatability and coherence with varying headphones. On the other hand there is still the attraction to replicate the natural working environment of the headphone - a human ear on a human head - as close as practically possible. Because as already pointed out, only this is capable of inducing the correct direction-dependent resonances that are part of the natural human hearing. But once again, without a reference system for comparison and the possibility to verify the specific benefit of physical changes of the rig it remains as kind of a poking around in the dark. There are really a lot of uncertainties going on in DIY. The only way to improve is probably measuring as much as possible, with varying headphones, positions and coupler modifications, then gather the data and compare.

What is actually producing the sound in the case above?

The device under test was a Shure KSE1500. The graph was posted by csglinux on head-fi:
https://www.head-fi.org/threads/audio-measurements-on-a-headfi-budget.893084/
His thread provides a great introduction into headphone measurement by the way.

does it matter or are we only interested in relative performance of the measuring devices?

The point of the measurement was to show up the resonances induced by the different rigs. Since the 60318-4 coupler is still widespread and being accepted as a good standard in the community the aim of this approach is to find a system that shows a similar performance to facilitate cross-comparability.

Measuring over-ear headphones is a differnt thing though because it extends the sound path, including ear canal and pinna.

The miniDSP EARS I critizied in one of my first posts could be accepted as an affordable comprise. As a mass-produced product it is easy to obtain, easy to use and calibrated in factory. It lacks compatability for on-ears and IEMs and won't reach the scientific credebility of B&K and GRAS rigs, of course. Still, it comes with a more or less known response and could thus be treated as a decent standard for hobbyists. I think Brent Butterworth from SoundStage.com found the right words for it:

As I see it, the “killer app” for the EARS is for over-ear headphone DIYers who physically modify their headphones in pursuit of better sound. With the EARS, they can easily (and, I think, accurately) gauge the results of their modifications relative to the unmodified version, thus saving themselves many hours of trial and error, and almost certainly achieving better end results.

If we look at it this way, rigs like the miniDSP EARS are probably totally fine as long as they stay consistent and are treated by a person with some knowledge about headphone measurement.

Looking back at my own project I cannot claim such a standard (not even an inofficial one) and a direct cross-comparability since my build is very specific at the moment. As a goal I try to find a good basis for finding diffferences between headphones and modding them. And if possible, tweak the build to give consistent results with peaks somewhat similar to the established standards.

Going the simplest and cheapest route, the flat plate coupler is probably the best way as long as you know what you are doing and can confirm decent repeatability.

I hope that clarifies a few things.

 

[Deleted member 16543]

It all depends on your goals and ambitions at the end. If you want to measure what you (physically) hear, there is no other way than starting with an anatomically correct replication of your own ears, indeed. Since that is hard to accomplish for the various reasons we already gathered, the best we can get is a system whose measurements have no absolute weight but allow comparative evaluations at least.

That's definitely one way to look at it. On the other hand, there's at least one instance of promising convergence of measurement results with the anatomically accurate way of doing things. It's also the only instance of comparison between two anatomically accurate ear model measurements.. so that's promising to me. This is a very new field. I try to focus on its potentials, while not dismissing the 'poking around in the dark' side of it. Just recently, a guy on DIY audio was able to reproduce the results of an experiment done with full ear replicas vs closed canal, done about 25 years ago, with a rig he totally made from scratch.
Just because the ear rig looks weird and unusual compared to tried and (un)true couplers or flat plates, it doesn't mean that the results have to be all over the place among different anatomically accurate models. There's going to be differences, but not more divergent than the ones from coupler to another coupler, for example. A comparison between the B&K and GRAS full ear models might be very informative in that regard. I doubt there will be great deviance between the two up to about 10 kHz, or even above.
If one is up to doing this test I will provide them a pair of my own ear models for free, as a third comparison rig. Just PM me. I can also ask the guy on DIY audio if he's willing to do the same.

Maybe it's just me, but I believe if things are worth being done, they're worth being done to the best of one's efforts. Not taking anything away from the 'simplest and cheapest' approach, which definitely has its own merits, sometimes.

 

[Earfonia]

Sorry for late reply, I've been quite busy during this pandemic season due to increase workload. And will be quite busy throughout this year.
I've bookmarked this thread so I can read thoroughly when I got time.

Thanks for the updates!

Hi Earfonia,
The Rode VXLR+ is a handy tool for adapting electret microphones to phantom power. I have build a similar adapter with the following circuitry: http://www.jp137.com/lts/XLR.PIP.pdf
Building the adapter myself allowed me to tune the load resistance. This gave me the chance to apply a higher drain-source voltage which increases the mic's sensitivity.

Thanks for the schematic!

(1.) EDIT:
Could it be that most of the measurement mics below $100 are actually the same?

Presonus PRM1:
- Sensitivity: 14 mV/Pa
- Impedance: 200 Ohm
- DR: 106 dB
- SNR: 70 dB
- Max SPL: 132 dB

Superlux ECM-999:
- Sensitivity: 14 mV/Pa
- Impedance: 200 Ohm
- DR: 108 dB
- SNR: 70 dB
- Max SPL: 132 dB

Sonarworks XREF 20:
- Sensitivity: 14 mV/Pa
- DR: 106 dB
- SNR: 70 dB
- Max SPL: 132 dB

t.bone MM-1:
- Impedance: 200 Ohm
- SNR: 70 dB
- Max SPL: 132 dB

Same housings and pretty much the same specs.

Wow very good observation! I have Superlux ECM888B:
https://www.aliexpress.com/item/32812063999.html?spm=a2g0s.9042311.0.0.27424c4dOL3lEP

I bought it because I curious with the flat FR on the spec sheet. But sadly haven't got time to play around with it. Used it a few times with DIY coupler to measure IEMs, but haven't compared it with other mics.

(2.) EDIT:
According to GRAS' specifications the 60318-4 coupler shall be used with a 40AG pressure microphone (PZM). Here is its frequency response graph:

View attachment 58284
https://www.gras.dk/products/measur...s-200-v/product/ss_export/pdf2?product_id=167

Looks like the referenced response in the spectrum below 20 kHz is most equal to the 90° configuration.
So I guess that a "consumer" microphone tuned for diffuse field measurements should be more or less sufficient. GRAS recommends a random incidence correction of -3 dB in the upper frequencies, though. See also p. 19 to p. 41 of this presentation: https://www.gras.dk/files/MiscFiles/News/GRAS Alma 2017 Micropones and couplers PWA 170102.pdf

AliExpress sells pressure microphones for IEC711 couplers, not cheap but probably good enough:
https://www.aliexpress.com/store/gr...967.html?spm=2114.12010615.0.0.6eb943e2ZYEnce

The cheapest version of IEC60318-4 coupler with mic cost around $100 each:
https://www.aliexpress.com/item/4000504988211.html?spm=2114.12010612.8148356.22.dd6d3bd0Ek1TVm
https://www.aliexpress.com/item/4000789796521.html?spm=2114.12010612.8148356.14.dd6d3bd0Ek1TVm

Probably worth checking

 

[Dreyfus]

Just recently, a guy on DIY audio was able to reproduce the results of an experiment done with full ear replicas vs closed canal, done about 25 years ago, with a rig he totally made from scratch.

Sounds interesting. Do you have a link to that?

Maybe it's just me, but I believe if things are worth being done, they're worth being done to the best of one's efforts. Not taking anything away from the 'simplest and cheapest' approach, which definitely has its own merits, sometimes.

If we all had the time and money, sure.

I bought it because I curious with the flat FR on the spec sheet.

The specs are very vague and can practically vary more than 6 dB between samples of the ongoing production. See this page for examples:
https://www.hifi-selbstbau.de/grund...0-1000-mikrofonkalibrierungen-eine-uebersicht
(Just ask in case you have questions about the details of the article, I can translate.)

To my knowledge the ECM-8000 is somewhat flat when measuring around 60°. It is rather diffuse field than free field calibrated.

AliExpress sells pressure microphones for IEC711 couplers, not cheap but probably good enough ...
Probably worth checking

I though about purchasing them when ordering the two couplers. But I was too sceptical about the technical parameters of those mics.

 

[Deleted member 16543]

Sounds interesting. Do you have a link to that?

Post #40 on
https://www.diyaudio.com/forums/ven...-ears-anatomically-accurate-ear-canals-4.html

 

[Dreyfus]

Thank you for the link.

I have already seen the data by Hammershøi and Møller being referenced by David Griesinger who concluded that "The directional dependence as measured at a blocked ear canal can be correct – But the timbre is so incorrect that our ability to perceive the true direction is frustrated".

I think the strategy of Hammershøi and Møller is to show that ...
A) the overall distribution of pressure gathered in the frequency response is quite consistent for the three speaker positions when you measure inside or directly infront of the ear canal of an individual.
B) the "[ear canal resonance] variation from subject to subject is rather high" whilst the ear canal entrance point produces the least standard error of directional dependency between the subjects.

Looking at the data though you can see that there indeed are some peak amplifications that can affect the coloration and imply a directional dependency of the ear canal for higher frequencies. The Q of those peak amplifications should be wide enough to perceptibly change the sound. In consequence I agree that the simulation of the (anatomically correct) ear canal can play a role for binaural recordings. As for headphone measurements the response is probably much less affected as long as the driver size and position stays consistent.

I more and more get the impression that you should either go for a sophisticated anatomically correct simulation of the ear canal OR just totally skip on the canal and just use the ear entrance point which would include less individual significance BUT produce more consistent results due to the lower directional dependency. Also, as Hammershøi and Møller note "the data do not tend to support the simple model of the ear canal". So it is probably better to skip on the ear canal simulation than using an oversimplified model which adds false coloration.

 

[Deleted member 16543]

I agree that overall there is some consistency between the closed ear canal and a position inside of it, at least compared to the higher inconsistency measured above 10 kHz.
However, the inconsistencies that do still happen where our ears are most sensitive (1-3 kHz) are still important, especially when making precise measurements.
So I don't think there's much value in using a rig that minimizes directional dependency. If there is directional dependency (and there is), then you want to detect it, not smooth it out.
Unless one is interested in measuring only one type of headphones, the driver size, position (and chamber shape) are going to be different. The different contributions of these factors should need to be measured accurately, not concealed.

While I agree that not everybody has all the time and money to do things right, if one has already invested the time and money to get a pinna replica, then I would suggest them to go the extra mile, for virtually no money and very little time more, and add a plastic tube canal replica. It won't be an anatomically accurate model, but it will get you somewhere in the ballpark of an accurate measurement. More than a closed ear canal, at least.
Years ago I used clear plastic tubes, and a couple small zip ties wrapped around it to simulate the ear canal bends.

 

[Dreyfus]

To be honest I doubt that you can get any reliable accuracy for such phenomenons in DIY anyway. The pinna I am using is not super detailed. Also the coupling without a correct ear canal replica might have its flaws. So I think - in this case - it is rather a good idea to smooth things out a bit.

This is the status quo by the way:

The yellow element represents the 3D printed canal extension shown in post #12. You can see that I modeled the shape so that it smoothly adapts to the 7.5 mm wide entrance of the 60318-4 coupler.

The canal has no bends at the moment, just a straight extension. It is roughly 18 mm deep, measuring from the tip of the tragus to the metal grid of coupler.

I might play around with some plastic tubes and study the changes of the response. Narrowing them with some zip ties is a clever idea - thanks for the tip!

 

[Deleted member 16543]

Great work!
I really like the interposing extension to smooth out the connection between pinna and canal. That's a step I though about doing myself, but then I decided to skip directly to modeling the whole thing.
The pinnae you're using look detailed enough to me.

I don't know how much you spent for the 2 couplers, but considering $50 for the tubes, zip ties, plastic filament and mannequin head, plus about $40 if you really wanted to buy another set of life size silicone ears, you have yourself a better measurement rig than any of the other options (except the anatomically accurate one) for under $100.
That's not including the capsules and preamps, but those were going to be there with the other options as well.

You mentioned Primo EM258 in post #12. The maximum SPL for that capsule is 115dB. That is enough for measuring headphones, since you don't need to use high volumes for that. Acoustics are a highly linear phenomenon and you will get the same amplitude and phase response at 70 dBSPL or 100 dBSPL. However, if you also intend to measure the distortion of the headphone driver as you approach higher and higher listening levels, you might want a capsule with higher SPL handling.
Especially if you also intend to use the binaural mic for music recording, and considering that the canal+pinna adds peaks of up to 15-25 dB to what you would measure with a regular SPL meter, that leaves you with 90 dBSPL as measured with an SPL meter. You get close enough to a piano and you're going to easily overdrive the capsules.
With EM215 (150 dBSPL) or EM302 (156 dBSPL) you wouldn't have to worry about that.

 

[Dreyfus]

The head + torso I will be using is already set.

(fancy name / acronym might follow)

I still have to figure out how to cut it open to get access to the mounting points (left and right ear), without making it look like a mess on the outside.

The tubes would have to adapt from roughly 10 mm x 13 mm at the end of the concha / ear canal entrance to the 7.5 mm entrance of the coupler. I could try to widen some with heat and pressure, then apply the zip ties for the first and second bend. Or simply use kind of a soft shrinking tube to model the whole thing from scratch.
3D-modeling and printing would be an option, as well. But I would need a good source model (scan) for that.

I originally choose the EM172 / EM258 for their great SNR, accessibility and price. I do also have some PUI 5024 HD laying around which have a similar EM172 performance. THD is still on the weak side, though. That is the biggest drawback of those capsules, indeed.
The Line Audio OM1 is still an option since it has probably the best THD in the lower price range. Only 0.5% at 133 dB! It is not so easy to adapt to the coupler, though. Like already mentioned, I would have to modify (cut and angle) the housing to make it fit.
Behringers ECM-8000 and its companions would cause the same trouble, and are technicaly not even on eye level with the Primos imo.
EM215 and EM302 are much better for THD of course. But they really lack SNR. That would require even higher reference levels.

 

[scott wurcer]

The head + torso I will be using is already set.



I originally choose the EM172 / EM258 for their great SNR, accessibility and price. I do also have some PUI 5024 HD laying around which have a similar EM172 performance. THD is still on the weak side, though.

Why not use the 3 wire versions of these capsules, the distortion improvement can be dramatic?

EDIT- I see the EM258 is incompatible with the 3 wire mod (so they say). It does look like it would need some heroic XActo surgery.

 

[Deleted member 16543]

I don't think the SNR is too bad (but definitely not great). It is usually referenced at 94 dBSPL, so with the pinna+canal resonances you should get an A-weighted increase and get to 50 dB even with the lower SNR one of the two. Headphone driver distortion will be much higher than that at most testing levels.
However, you might want to get in touch with Primo and see if they still sell EM194s. 140 dBSPL and 68 db SNR. They are balanced though, so they require a balanced preamp.

PS. Nice head + torso. Is it actually real dimensions or a little smaller, like many unfortunately are?

 

[Dreyfus]

@scott wurcer:
Thanks for the reminder!
Last time I checked the market the EM173 was end of live and just out of stock at Micbooster. They now offer the successor EM273, though. That would be an option, indeed!

I have seen the 2-wire to 3-wire mods on micbuilders already. I resign.

@sax512:
Not bad, but not great either. I just try to get the best performance out of my budget to max out the usability of the rig.
Primo appears to be very picky with their vendors. And as far as I know they do not deliver smaller quantities for private demands.

The head is bit smaller than average. The coupling area is somewhat flat, though. Works much better for sealing than the other ones I tried in the past.
Maybe not the best for binaural recordings and HRTF purposes. But good for headphone measurements.

 

[Dreyfus]

Some news:
I took the ear canal pieces of the impressions I got in march (which hopefully didn't shrink too much) and inserted them into the silicon ears. The fit is actually quite good. I then modeled the extensions by just wrapping some Sugru around the alginate.




Here you can see how the DT 880 BE responds to it:

The ear canal adds some resonance around 2 kHz with the coupler and around 3.4 kHz without the coupler. If I compare that to my personal equal loudness equalization which falls between the blue and the yellow line I fear that this extension does not add any value for me. Compared to the old measurements this series is further away from what I actually hear. Maybe that is due to the Sugru being too hard and reflective. Maybe because the flange was flawed somehow. Or it is just the fact that I am trying to combine a random ear with a personal canal.

Maybe I will redo the procedure with some softer silicone. I tend to favour the symmetrical canal extension (blue line) shown in post #50, however.

 

[Deleted member 16543]

As usual, great work!
Your dedication is quite admirable to me.

I went and dug up some old graphs from 7 years ago on my back up hard drive. The one you might be interested in is this one.

This represents the FR of my binaural microphone, when a speaker is playing a white noise signal from a standard stereo configuration. Only one speaker is playing, the one on the same side of the capsule being recorded. The other speaker is off. One has to consider that this is not exactly the curve you should get from your headphones measurement that would indicate a perfectly balanced response.
The main deviations from a correct measurement are:

1. The fact that the speakers I used hadn't been equalized to a balanced target curve (but they probably weren't too terrible either).
2. The fact that this measurement doesn't take psychoachoustics principles into account at all.

However, indicatively (which is the only way you should take this graph!), you can see that there are two separate peaks at roughly the frequency you're getting with your pinna+canal rig. So I would say you are indeed on the right track with that setup.
What you would need to figure out now is what a response from balanced speakers measures like (fix the deviations above mentioned).
If your headphones measure close to that, then they are balanced headphones. If not, you can calculate the EQ difference and apply it to the signal, to make them sound balanced.

 

[Dreyfus]

Thank you! It is also a pleasure to get such great feedback from other experienced users.

At this point I think I do not have enough comparative data to conclusively judge the newer responses. All I can say is that the "pinna + coupler" curve is the closest to my preferred listening curve which is mostly flat up to about 1 kHz where the ear resonances start to rise. For me that is a practical thing since it requires less compensation when I want to look at the tonal balance.

If we follow the findings by Edgar Shaw there should be some resonance around 2.5 kHz for the ear canal and ear drum and around 5.5 kHz for the concha gain.

This does roughly align with the plot above. The 2 kHz resonance would maybe shift up a bit in frequency if I further extended the canal. My impressions covers only 3/4 of the total ear canal length at the moment.

We still do not know the inherent response of the DT 880 BE, though. The only measurements I could find are the flat plate ones done by Solderdude and the 43AG ones done by Oratory.
The GRAS reference holds a simple cylindrical 8 mm extension between the pinna and the 60318-4 coupler. No ear canal.
https://www.dropbox.com/s/ek8baed3t6fy6rd/Beyerdynamic DT880 Black Edition.pdf?dl=0
I am not sure what exactly the 3.5 kHz peak is related to. We can see quite a good consistency of the concha gain around 5.5 kHz across all those measurements, though.

I cannot do any speaker comparisons until the whole dummy head is constructed, unfortunately. Still a few things left to do (cut it in half, prepare the mounting points for the ears, add some absorption and damping for the inside, build the mics, ...).

We also have to keep in mind that the mic I used is not calibrated, yet. There should be some pressure build up followed by a roll off in the upper frequencies.

 

[Deleted member 16543]

Why do you want the measurement to match your preferred EQ (if that's what you mean by preferred listening curve)?
If anything, if you like to, say, bump up the mids with a certain model of headphones, it is because they lack mids to begin with. So your headphone should measure as somewhat lacking mids, not hyping them. They should measure as the specular of what your preferred EQ is. Specular in respect of the response of a 'perfect' pair of headphones, which we don't know what it looks like yet (but we'll get there, eventually).

If your canal replica is only 3/4 of the real length, you should try to find a way to add the missing part to it. No resonator is necessary, when you have a roughly accurate ear canal. It will only distort the geometry boundaries.
Lengthening the canal to proper size will lower the resonance peak though, not increase it. Which would make the peak from ear+canal get closer to the expected value. This is another indirect validation of the ear+canal system, if you think about it.. Although I wouldn't trust prediction models too much. The peaks depend on the source location and distribution in space, after all. It's the whole concept at the base of HRTFs..

 

[Dreyfus]

Do any of you happen to know an online source for soft silicone sheets?I am looking for some that I could use around the ear to extend the coupling surface for the ear pads. I found some white ones with 2 mm thickness on eBay and Ali. But they are at least 60 Shore A.
A little bit too hard for my taste.

Why do you want the measurement to match your preferred EQ (if that's what you mean by preferred listening curve)?

Personal choice for convencience. I could achieve the same result by adding more compensation to the other measurements. Or just choose the one that gives the most flat response when a (subjectively) flat headphone is measured straight away.

No resonator is necessary, when you have a roughly accurate ear canal.

What kind of resonator do you mean?