DIY Headphone Measurements - Ear Simulation?

[Dreyfus]

Thanks for your explinations.
I'm already familiar with the background of binaural (dummy head) recording. And I think you are totally right on most aspects.

As far as I know the GRAS ears can be ordered via their website or specialized distributors in your country. I haven't required a quote, yet. But as far as I know their business model the price will be out of the league for ordinary mortals. Especially for the newest generation of anthropomorphic pinnas with enhanced ear canals.

The pair of ears I will use is shown above. It is a symmetrical pair of soft silicone ears. The size is just right with 67 mm top to bottom lobe. Some other models I got in the past were much to large and had trouble fitting inside some ear cups. I do still have some concerns about the 3D-printed adapter and the resonances it might introduce. Still, probably better than simply sticking the ear simulator right to the ear canal entrance. Afaik the 60318-4 is designed to be coupled close to the ear drum reference point (where an in-situ microphone would sit).

As you can see I also tried molding my own personalized pair of silicon ears. Unfortunately the compound the audiologist used was not that optimal for detailed impressions of the human ear. And even if it was, further processing, shaping and molding the replica is really an art on its own. Not so great if you don't have the practical experience and technical equipment (like a vacuum chamber for degassing).
This is why I will probably stick with the model shown above. It might be kind of random and a bit simplified in its shape. But still better than simply measuring on a flat coupler without any pinna interaction.

I know that - after all that - it is nearly impossible to hit the same baseline as the industry standards do. But I will try my best to come at least close to it, as reproducable as possible.

The mics I consider at the moment are the Dayton iMM-6 and the Superlux ECM 888. They are not the most sophisticated measurement mics, for sure. But with some calibration they will probably do the job. I will definitely have to apply some compensation for the 0° incidence in case I want to get near to the reference baseline. As I already suggested above, ear simulators are tuned for pressure response. For most measuring mics this response will be most equal to the 90° incidence.

Noise floor as a parameter of dynamic range will play a role when I want to measure things like crossfeed. Distortion measurements would be great as well. But that's where the cheap capsules and their preamps really limit. The best I could find within my price range is the Line Audio OM1 with 0.5% at 133 dB. Sadly it will not fit the coupler due to its nozzle.
The Primos are great in price and performance, especially for their 80 dB SNR. High SPL is a weakspot, though. Even with advanced circuitry and 3-wire hacks.

Regards,
P48

 

[Deleted member 16543]

Yeah, if you want to measure crossfeed then you need something not too noisy. However, with some measurement software such as Acourate, with sine sweep type of test signal, the S/N ratio is increased a lot compared to, say, white noise. I think 120-125 dBSPL handling should be enough for any realistic listening level measurement, but that Line Audio mic does look very good!
Have you ever seen these while researching? I made them about 3 years ago.. https://www.shapeways.com/shops/binaural-microphone-ears. But they won't fit the OM1 capsules.

 

[Inner Space]

It seems to me that your design will be as good as anything else as a binaural mike. But for headphone testing, I'm not sure. I was part of some abandoned research that showed (with over-ears) that bone conduction makes a significant contribution to perceived SQ. No one knew how to record it, measure it, or evaluate it, which was why the program ended.

 

[Dreyfus]

@sax512:
Indeed! I think I have seen your posts on Gearslutz before.
But to be honest, non-elastic plastic is a no-go for me.

I plan to use Virtins Multi-Instrument or Room EQ Wizard for the measurements.

@Inner Space:
That might be a hot debate. And I fear to touch it.

 

[Deleted member 16543]

Very interesting, @Inner Space
It makes sense to me that, with headphones, the direct transmission of mechanical vibrations requires them to be transmitted through a medium that is more closely matched to a real human head. That's a lesser order component when it comes to binaural recordings, but probably not to be disregarded with headphones. The B&K ears seem to have a combination of solid and soft material that might (but then again might very possibly not) approximate what you're saying.

@P48 Because of the above, you're probably right to not want solid materials. However, just because a material is soft, it doesn't mean it's acoustically similar to flesh. I wouldn't put my hand on fire that even ballistic gel comes that close to that. Not before seeing some serious study on it, at least. Anyway, I definitely appreciate your purist approach. It's nice to talk to people that want to be even more specific than yourself and particular about the details, as opposed to the majority of people on Gearslutz

 

[Dreyfus]

I haven't seen studies with different materials, yet. But I suspect that plastic will have totally different reflection and absorption properties than a real human ear. The properties of silicone come much closer to that imo. It can also deform when under stress which is important for sealing with on ear headphones or over ears with smaller cups.
To be fair, the 3D-printed adapter between the pinna and the ear simulator is made of hard plastic, as well. Even the official adapters offered by GRAS are made of hard steel and some actually of plastic, though. As already mentioned, only the newer models come with ear canals molded with silicone.

Thanks for the appreciation btw! I can only return that compliment. Your posts were a valuable inspirations on my journey.
Nice to see another "ears guy" on the ASR forums!

 

[Deleted member 16543]

Oh I'm a speakers guy now. But still enjoy talking about this stuff.
I'm glad I offered something to reflect on with my posts, in the past. I'm interested to see where this branch of research ends up going. It seems such a niche interest as far as recordings go. Hopefully the headphone measurement side of it will pick up the slack.

 

[Dreyfus]

Hello everyone,
I have some news about my project and the topic of ear simulation after playing around with some prototypes.

I tried to measure the resonances induced by:
- a flat plate coupler
- a plate with only a silicon ear attached
- a plate with ear, a short canal extension and a 60318-4 simulator attached

The prototypes were made quick and dirty with some cheap plywood with a hole cut through. As a mic I used the PUI 5024HD electret which was fixed to the hole and sealed around with Newplast. The contact surfaces where the pad of the headpone sit was covered with some vinyl film.

The improvised setups were then used to capture the response of a headphone I had just laying around, the Beyerdynamic DT 880 Black Edition.
Here you can see the graphs plotted in REW, matched in level to 100 Hz and 1000 Hz:

Note: Please ignore the ripple in the lower spectrum. That's due to me pressing the cup against the wooden plate by hand.

My assumption is that the bass stays consistent when a good seal is applied to both the mic and headphone and that the change in coupling does predominantly affect the higher frequiences, starting somewhere around 200 Hz. This is also backed up by scientific investigations that demonstrate the influence of the pinna, ear canal, neck, torso, etc. See The External Ear by Edgar Shaw from 1974 as an example. Thus the best way to compare the effects of the setups introduced above should be the upper graph which is normalized to 100 Hz.

Now, starting from there you can clearly see the varying slopes that occur when measuring through the builds. The blue line which is the flat plate coupler with the mic right at the boundary layer shows the least resonance in the upper mids and above due to the lack of ear impedance whilst the red line with the pinna, the short canal extension (as described in post #12) and the 60318-4 ear simulator are raised by up to 13 dB.

But what do all of those graphs actually tell us about the tonality and balance of the headphone?

I recently debated with @solderdude who takes the view that a headphone should be flat when none of the effects above are 'distorting' the sound so you can look at the response of (more or less) only the transducer. I took that as an inspiration and generated some EQ curves with the 100 Hz and 1 kHz reference line to hear how the different setups would sound when equalized to flat from 20 Hz to 4 kHz.

As you would expect, the flat plate has the strongest discrepancy with the bass and presence region being up to 11 dB apart from each other. Listening to both white noise and music with the DT 880 Black Edition I noticed that there was a fundamental lack of bass or - to consider it the other way - a severe overemphasis on the presence region. Far from neutral, if you ask me.
The next one, which is the pinna only setup, is much more appealing. It seperates the the upper bass / lower mids and presence region by only
7 dB and doesn't make the Beyer sound nearly as thin and tinny. There is still some emphasis on the upper presence. Playing a few rounds of Call of Duty Warzone however I noticed that this curve is actually quite similar to the one I got used to the past few months. The timbre up to 4 kHz is close to that of my modded HD 700 which I matched for equal loudness half a year ago.
And finally the pinna with ear simulator setup which has about 7 dB of correction as well but adds an attenuation between 1 kHz and 4 kHz. This makes the headphone sound less forward and a bit muffled. I would say that the balance is probably somewhere between the pinna only and pinna with pinna with ear simulator setup, with a tendency to the first one which is also closer to the curve offered by Oratory.

So far, here are my thoughts and conclusions on all that ...

1) The flat plate is too sloped and cannot be considered to sound balanced when the headphone is measuring flat between 20 Hz and 4 kHz.
2) The pinna only and pinna with ear simulator setups can be considered to sound fairly balanced when measuring flat until 4 kHz.
3) All of the observations described above may change to some extent with varying damping around the mic and ear. I improvised the boundary layer with some vinyl but may try thin layers of silicone, rubber or closed cell foam.
4) You could debate whether or not the response should actually be flat until 4 kHz. The elevation towards the higher frequencies depends on the azimuth and actual spetral map you would consider to sound flat, of course. You could also include the response of a specific room, averaged listener preferences and all that ...
I'm trying to not overthink that and just listen to my own ears. The upper limit of 4 kHz as reference for a balanced sound may be a bit too high. Still, the experiments show that the quick builds made with pinna and ear simulator can provide reasonable results. Now I can try further modifications and survey how well they improve the response in relation to what my own ears perceive. I do also have to test the repeatability of my findings with different headphones.
5) Treble is still an issue and cannot be accurately measured with such a rig. Even the "standards" by Oratory or rtings cannot fulfill that, if I compare the equalizations derived from those with my own preception. So I will probably ignore all the sharp resonances above 4 kHz and just look for the overall boost or attenuation in this region. Further investigations follow.
6) Bass will be individually compensated, of course. I have to make sure that the seal stays consistent with differnet headphones, though.
7) The mic is not calibrated, yet. There might be some inconsistencies in the upper frequencies due to the inherent sensitivity and the pressure build up infront of the capsule. I probably have to recheck the findings above after calibration.
8) Still thinking about whether or not I should go for a dummy head for this project. I may just 3D print a base unit and attach my ears and couplers to those. We will see ...


To be continued.

Regards
P48

 

[solderdude]

Strangely enough my flatplate measures about the same as your Pinna, except for the treble area.

Try the following to calibrate the mic:
Measure a speaker with the mic in free-air
Measure a speaker at the same distance while mounted on the baffle.

Here is how I measure the DT880 Black SE on a flatbed (but not simply a not calibrated mic on a baffle)

The DT880 (original) has, just like the HD650, very little Pinna activation, certainly not 10dB.

Question.. does the Pinna push against the driver ? Its why Tyll got incorrect measurements sometimes

You can use a HD650 or HD600 to get a proper calibration.

 

[Deleted member 16543]

Great job, P48!
Painstaking work and outstanding presentation.

As far as the calibration goes, you may want to start by looking into the latest target curve research for speakers. In short, for speakers, a response is more or less balanced (the actual result depends on the recording, of course), when it measures flat from 20 Hz to 1 kHz and then slopes constantly from there to -6dB @ 20 kHz.
The gating of the IR measurement is also very important. Not only there is no such thing as a speaker measurement that can be meaningful everywhere in the room, but the measurement at the listening position needs to be made according to psycho-acoustic considerations to make it translate into effective sound quality attributes. See this book, which I can't recommend enough:
https://www.amazon.com/gp/product/B01FURPS40/ref=ppx_yo_dt_b_search_asin_title?ie=UTF8&psc=1

I checked that this curve is not only valid for speakers in a standard room, with the usual mix of direct and reflected sound, but also -according to my ears at least, and I did try the flat target- with special DIY speakers I made, designed to minimize the room contribution at the listening position (very near field speakers).

All of the above to say that if we can assume, as my experiments seem to indicate, that the correct balance is sloped and independent from the type of listening position (far vs near field) with speakers, and be willing (reasonably, but with need of confirmation) to expect the same with headphone listening (which is also the philosophy behind headphones such as PSB and NAD), then an expectation of flatness is not necessarily a given anywhere in the equalization process.

This is the most logic driven calibration procedure for headphones I can think of (all measurements done with appropriate psycho-acoustic gating considerations):

1. Sit at the listening position of speakers equalized (with calibrated measurement mic) to match the target curve indicated in the book I linked to.
2. Find the equal loudness equalization with those speakers.
3. Put the headphones on (making sure DAC and preamp are flat, otherwise their deviations must be taken into account as well).
4. Find the headphone equal loudness equalization.
5. Correct the headphones equal loudness EQ with the inverse of the speaker equal loudness equalization. This should make the headphones (on first approximation, but without HRTF considerations due to different positions of speakers and headphones in respect to the listener's eardrums) sound more or less as balanced as the speakers. I would do this step with a linear phase filter, but minimum phase comparison could be interesting to consider too.
6. Measure the headphones response with a rig of your choosing and compare. This will give you the target response to aim for (short of the mic calibration, of course)

Alternatively, a variation on the procedure by David Griesinger could be done. What I don't like about that procedure is that it requires to put microphone capsules close to the eardrums. This is not ideal, not only because of geometry considerations, but for safety ones.

So, unless one wants to specifically equalize for just themselves, an average ear canal replica is necessary for the most meaningful results to be extracted from this procedure. This is the same principle I applied to my binaural microphone design. Average ear canal to work for the majority of people or personalized ear canal for optimized but more individual results? Hard to pick. I made my choice but it comes with its pros and cons, just like anything in audio.
Anyway, the reason why you need a canal (average or personalized) is that you already have the HRTF considerations above mentioned to take into account. You wouldn't want to add unnecessary microphone related transfer functions to the equation, if possible. On the other hand, just like with the phase response of the correction filter on point 5, as an investigative research, there's nothing wrong in trying all possible configurations and see which one comes closer to a good balanced sound.

 

[Dreyfus]

Thank you for the great feedback!

Strangely enough my flatplate measures about the same as your Pinna, except for the treble area.

Interesting!
May I ask what kind of damping you used?

Measure a speaker with the mic in free-air
Measure a speaker at the same distance while mounted on the baffle.

Do you use this kind of calibration for all your plots?

The measurements I made were completely raw, no compensation applied. Calibrating the mic for the pressure response is probably the way to go. Afaik that is also the reference for the ear simulator.

The DT880 (original) has, just like the HD650, very little Pinna activation, certainly not 10dB.

Thanks for the hint, good to know!

Question.. does the Pinna push against the driver ? Its why Tyll got incorrect measurements sometimes.

I just checked that with roughly the same pressure I applied for the measurements above. I would say there is about 5 mm of compression on the pinna. The silicon ear I use is quite soft, though.

I will keep that in mind for further measurements and try to reduce the pressure (which was rather high so far, pressing the pads against the plate by hand).

As far as the calibration goes, you may want to start by looking into the latest target curve research for speakers. In short, for speakers, a response is more or less balanced (the actual result depends on the recording, of course), when it measures flat from 20 Hz to 1 kHz and then slopes constantly from there to -6dB @ 20 kHz.

Thanks for the hint! Is this linked to a certain study?

All of the above to say that if we can assume, as my experiments seem to indicate, that the correct balance is sloped and independent from the type of listening position (far vs near field) with speakers, and be willing (reasonably, but with need of confirmation) to expect the same with headphone listening (which is also the philosophy behind headphones such as PSB and NAD), then an expectation of flatness is not necessarily a given anywhere in the equalization process.

That is a tough topic!
Finding a good reference curve for headphones is a thread on its own.

When it comes to speakers in a room I have no strong opinion on how the actual response at the listening position should be because I'm a 95% headphone guy and use a subjective curve with no specific "external" reference. So far I just calibrated my desktop speakers for the Harman house curve and attenuated the LF boost if neccessary.

Maybe I will find the time to do some speaker comparisons with all the mentioned approaches any time soon. I don't want to get my head too deep down the rabbit hole though. My initial aim was to find a target that works well with my personal headphone equal loudness preference of the past few months.

I think the next step is to get my mic calibrated. My Beyer MM1 is already a few years old. But I will get it checked in a few weeks for its 0° and 90° response. That should give me a good reference for matching the PUI capsules.

Average ear canal to work for the majority of people or personalized ear canal for optimized but more individual results? Hard to pick. I made my choice but it comes with its pros and cons, just like anything in audio.

I know that conflict too well!
After all I guess I will just use a simplified canal extension like shown in post #12.

... there's nothing wrong in trying all possible configurations and see which one comes closer to a good balanced sound

This!

Best wishes,
P48

 

[Deleted member 16543]

Several studies are linked in that book. Other target responses are discussed as well. Of course there is no ONE target curve everybody agrees upon, but flat at the listening location is definitely too 'clinical'.
I would be surprised if, if ever you decide to equalize your speakers, your own target response would end up being considerably different from the one I described.
The only reason I mentioned speakers is because at least there is a (somewhat) agreed upon target response.
Using that as a starting point, and the procedure described, you should get to your headphone target curve in the most scientific way possible.
The only caveat in that procedure is that the EQ to go from speaker equal loudness to target response at the listening position, measured with a mic in free space, is not necessarily the same EQ that it takes from going to an equal loudness headphone response to something that sounds as balanced as the speakers do when calibrated (HRTF considerations I was saying before). However, it should also be not that far off, especially considering that the psycho-acoustic equalization procedure I'm talking about largely disregards cancellation related narrow dips.
A possible variation of that procedure (and sorry if I start with speakers again, but I really don't see any way around it) could be to equalize the speakers, then measure the response of a binaural microphone with ear canals placed at the listening position, where the speaker measuring mic was.
That's your target response, achieved in the straightest way possible. I just thought I'd add some degree of personalization in the previous procedure, by adding the equal loudness steps.

 

[solderdude]

Interesting!
May I ask what kind of damping you used?

3mm thick cell foam rubber on 18mm thick wood small mic WM61A capsule.

Do you use this kind of calibration for all your plots?

Same compensation for all plots. The lack of Pinna results in small differences in the 2 to 5kHz range with some headphones, somewhat larger in others. I always check Rtings Pinna effect plots if possible. They are the only ones testing this.

My compensation is based on differences between 80dB and 70dB SPL equal loudness contours and the mic has it's treble peak removed which happens when baffle mounted in close proximity to the source.
You can check this with white noise on a tweeter 0.5m away and moving it closer.
Also when the mic is free air and baffle mounted. Should be less of a problem with omni mics.

 

[Deleted member 16543]

Should be less of a problem with omni mics.

I wouldn't even consider any other type of microphone, other than small omni.

 

[Dreyfus]

Thanks for the replies!

@sax512:
The comparison versus a speaker makes sense. I will look into the book you linked when I find the time and do some further experiments.

@solderdude:
The treble peak is what I was refering to mentioning the pressure build up infront of the capsule. The more you concentrate the sound field towards the 0° on-axis the stronger the boost, up to 6 dB in general. This effect can be masked with increasing diffusion which is probably the strongest with the pinna + canal extension + ear simulator described above. Hard to tell which kind of compensation is needed with such rigs, though. The effect will probably also vary with the size and angle of the driver, the damping around the baffle, the diffraction caused by the tragus (if present) and such.

I think I will start with the pressure zone calibration, having the capsule mounted in a thick wooden panel of at least 1 x 1 m in size. To minimize reflections I could mount the whole thing up in the air with tripods somewhere in the garden. That would be (more or less) similar to measuring speaker drivers in FF.

Regards
P48 (who will be out of town for a few days)

 

[Deleted member 16543]

I'm not sure that removing the peak is a great idea. The peak is most likely the result of constructive reflections bouncing back and forth from the driver to the baffle due to closeness of the source. By the way, I would also expect cancellation dips for the same reason.
Anyway, closeness of source is an inherent condition of headphone listening, so the same principle of constructive/destructive reflections apply during listening. That's why removing the peak might take you further away from an actual relation to the perceived sound.
However, that peak is not the same peak you get at your eardrum when you listen to headphones. That's why it's important to mimic the actual response at the eardrum by preserving realistic geometric boundaries as much as possible.
The flat baffle approach is probably good enough for a rough estimate and for rough comparison of headphone A vs B sound signature, but I wouldn't use it to try and actually equalize headphones to a target response, whatever that response might be.

 

[Deleted member 16543]

I had an interesting discussion about this on Facebook (of all places!) the past couple days. Somebody shared these two videos that I think might be interesting to you guys..
The research leaves something to be desired. Namely, the lack of phase response correction, the too approximate geometry of the ear models (no ear canals) and no description of how exactly speakers and headphones FR was calculated. Nonetheless it's a very interesting source, in my opinion.

 

[Deleted member 16543]

As I'm waiting on some parts to move my speaker design project towards completion, and since I guess I've never really gotten over my first love for binaural recordings and microphones, I started to do some research on closely related subjects and I bumped into a link that you guys might find interesting. As it turns out headphones measurement is very closely related to binaural microphones, at least in my opinion.
So I thought I would write a recap post to share my views on the matter of headphones measurement.

This is the link, which ranks headphones according to their deviation from the Harman target curve.

https://github.com/jaakkopasanen/Au...oo8qM2b-JZHyqcxYwSN73-7Vci2yisJSkTh_zFS7jHG1M


I have found no specific detail on Harman's measuring apparatus yet. The videos I linked to briefly show ears with no ear canals.
Not too surprisingly, the Harman target curve closely matches the measurements I was getting when equalizing the earliest revision of my DIY binaural microphone (when I didn't add the ear canal, yet). The problem with that (closed ear canal), in the case of binaural recordings, is that the effect of the ear canal (therefore its imprint on the sound waves actually sensed by our own eardrums) is direction dependent. So, if using closed ear canals when you have real events you want to record (with sounds coming from all directions, including 360 deg room reflections), you are effectively applying a direction dependent EQ on the recorded sound. That detracts from the realism of the recording.
In the case of headphones measurement there is no direction dependency, as the sound source -the headphone driver- is more or less fixed in space. But the effects of constructive/destructive reflections now are way more amplified in respect to the binaural recordings case, because of the closeness of the source to the eardrum and the tight quarters geometry. Modifying the geometry too much shifts the peaks and dips away from what the eardrum senses when one is actually listening to the headphones. So, to me, the closed ear canal seems to be a flawed approach for headphones measurement too, just for different reasons than accuracy in binaural recordings.

Anyway, even with the closed ear canal approach, one can see that the target curve is far from flat. There is a 8 dB bump around 3 kHz, for one. I'm not saying that the Harman target is necessarily the Holy Grail of headphone response, for the reasons above mentioned. However, that is good enough proof for me that looking for/equalizing to flatness is a flawed approach, when dealing with headphones.

In conclusion, if I wanted to seriously get into the headphone's measurement business, I would buy one (or a pair) of those ear replicas which include ear canals -to my knowledge at this moment available only from two reputable companies (B&K and GRAS)- and use a dummy head to try to recreate as close as possible a realistic headphone wearing condition.
Then put the dummy head at the listening position of a well balanced sounding speakers setup and measure the response at the eardrum (capsule).
That's the target curve one should be looking for.
Deviations from this recipe in the direction of simplification, keeping in mind that we are talking about scientific measurements and not artistic binaural recordings, is to be considered substandard in my opinion, given today's state of the art of readily available measuring apparatuses.
Of course, deviations in the direction of even more accuracy (like bone and flesh replica) is a different matter. But I think that at a certain point, given the necessarily approximate nature of the target response and variance in measurements due to headphones seating during tests, increased ear model accuracy will not give you more precise results. Where that point is with headphones measurement I'm not sure.
In the case of binaural microphones some folks are perfectly happy with closed ear canal, but that's subjectivity vs scientific measurements we're talking about.

Next time I get around to re-equalize my binaural microphone with something more accurate than Logic MatchEQ I will share those amplitude and phase responses.
Those responses will most likely be:
1. Measured using my own DIY binaural microphone (with anatomically accurate ear canals, but not from those reputable companies)
2. With said microphone placed at the listening position of a setup using speakers that have been equalized to a flat curve up to 1 kHz and sloping down to -6dB @ 20 kHz
3. Gated according to the psycho-acoustic algorithm used in Acourate (both for the speaker equalization and the eardrum response).

 

[Dreyfus]

I'm not sure that removing the peak is a great idea. The peak is most likely the result of constructive reflections bouncing back and forth from the driver to the baffle due to closeness of the source. By the way, I would also expect cancellation dips for the same reason.

That's a valid objection. My 5" speakers are not spherical and may introduce some errors due to reflections against the wooden wall. Maybe I can get that under control with some narrow gating?

Alternatively I could match the mic to a known reference in nearfield without the baffle in place. I could get my Beyer MM1 calibrated by a third party for 0° and 90° and then compare the response of that to my Primo / PUI capsules. However, such a calibration would only be valid for measurements without boundaries. Having the mic inserted into a baffle (flat plate) or ear simulator would require the pressure response, as already mentioned.

Another option would be to purchase a mic calibrated for pressure response. I found a supplier that offers some Behringer ECM-8000 which have been measured in a pressure chamber. Taking that as a reference I could insert the mics into the ear simulator and then measure the response of an IEM to find the compensation for my electret capsules. That would be similar to the reccomendation by Earfonia in post #15.

I had an interesting discussion about this on Facebook (of all places!) the past couple days. Somebody shared these two videos that I think might be interesting to you guys..
The research leaves something to be desired. Namely, the lack of phase response correction, the too approximate geometry of the ear models (no ear canals) and no description of how exactly speakers and headphones FR was calculated. Nonetheless it's a very interesting source, in my opinion.

Thanks for the links! That's a good refresh to my knowledge of harman research.

I have found no specific detail on Harman's measuring apparatus yet. The videos I linked to briefly show ears with no ear canals.

The 2017 presentation shows that they used the GRAS 45 CA with a modified IEC pinna (bass coupling optimizations). They probably did some experiments based on the newer anthropometric pinna with the extended ear canal (including first and second bend) in the meantime. Maybe someone with access to the AES papers can check for that?

 

[Deleted member 16543]

That's a valid objection. My 5" speakers are not spherical and may introduce some errors due to reflections against the wooden wall. Maybe I can get that under control with some narrow gating?

It depends on the distance from speaker to microphone (and the speaker's directivity as well). The closer the distance, the more the speaker driver itself becomes what's part of the reflective interference mechanism. Farther away, the boundary gain from a flat surface surrounding the capsule becomes more predominant in the measured signal. Like those mics you put on a surface instead of a stand.

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.
The more the geometry is different, the farther from a meaningful measurement you get.

The gating changes results too, that's why doing it according to psycho-acoustics principles is just as important as the measuring apparatus geometry (and material, if you think that is important too).