I'm guessing there is a practical reason that I'm not seeing, but it seems like one could design the worlds best headphone by aiming for a dead flat response with very little distortion(just like this HP), and then applying the HRTF via DSP.
If we're strictly speaking of HRTF I can think of several issues :
- if proven that generic HRTFs can’t be effective at sound localisation, gathering the anatomical data to turn it into an individualised HRTF profile.
- headphones measuring differently on your head than on a third party test rig for various reasons (lateral and vertical positioning variations, clamping force / pad compression / pad wear, the anatomy of your ear, etc.).
In regards to the former we’ve already seen quite a few attempts at using visual cues to derive individualised HRTFs, such as using user-generated photos or videos (Sony, Genelec unless I’m mistaken ?). I’m not sure that this provides enough data points to work effectively. But given enough data algorithms seem quite adept at generating individualised HRTF profiles (there are countless articles on the subject,I’ve also seen a few videos on the subject such as this one :
)
Here I think that Apple is in quite a unique position. With Face ID on their iPhones they can already gather quite a bit of data on the user’s head and perhaps the system could be used more accurately than a simply camera system for photos of one’s ear. But maybe even better, perhaps Apple could embed the sensors within the earcups themselves.
Quite a few of Apple’s patents released in the four years prior to the APM’s release show that they envisioned the use of various types of sensors to make a more or less precise image of the user’s ears within the cups. The most detailed patents mention the use of capacitance sensors for fairly simple applications (simply recognising the rough shape of the ear to make the headphones left / right reversible for example). For example :
https://appft1.uspto.gov/netacgi/nph-Parser?Sect1=PTO1&Sect2=HITOFF&d=PG01&p=1&u=/netahtml/PTO/srchnum.html&r=1&f=G&l=50&s1="20200280785".PGNR.&OS=DN/20200280785&RS=DN/20200280785
Unfortunately none of that happened in the final product
. But perhaps in the future we may see an increasing amount of sensors within the earcups and headphones may be able to provide an increasingly higher res image of our ears.
In regards to the latter the DIY probe / tube microphone that I’m making has been quite helpful in characterising this issue above the range where in concha microphones (of this type for example :
https://www.soundprofessionals.com/cgi-bin/gold/item/SP-TFB-2) start to become useless. It’s based on this video from David Griesinger :
. Although at this point it deviates quite a bit from it.
I’m now starting to get quite confident about using it for comparative evaluations up to around 10kHz. During the same measurements session I like to bookend it by measuring a pair of known headphones with a low positional variation to make sure that the probe didn’t move to any signifiant degree during the session. This is the typical sort of variation that you’ll see, here with a HD650, normalised at 505Hz (so basically within 1dB and the variation at higher frequencies is quite linear to begin with), between the very beginning and the end of a 2 hours session, after god knows how many headphones changes and seatings :
Across different sessions (on a different day) the variation tends to be higher than I’d like, essentially because of the probe’s position within my ear canal. Here is a difference between a session yesterday and one a couple weeks ago, still with the HD650 :
I’m getting better at consistently locating the probe but it’s a work in progress.
That being said, in different sessions, regardless of the exact absolute numbers, comparing my HD650 to my HD560S yielded very similar
comparative results (HD560S in blue, HD650 in red), two different sessions a week appart :
So I think that it’s a pretty solid method for comparative evaluations, regardless of the probe’s exact position within my ear canal. It's not rock solid mind you, and certainly should not be used to assert things such as "on my head, headphones X has 3.765dB more than headphones Y at 4327.3Hz", but just to assess trends in a somewhat loose way ("at 7300Hz or so, headphones X tend to be 2.5-3dB higher than headphones Y").
So, in regards to how headphones vary on your head vs. test rigs measurements, something quite simple to do, since comparative results are quite effective with this probe, simply is to apply Oratory1990’s (or others) presets and measure. This is how the HD650 (red) and HD560S (blue) measure on my head with Oratory’s Harman presets, first with the DIY probe mics and then with in-concha mics (valid for
comparative results up to around 3 kHz or so with these large open over-ears ?) - solid traces with Oratory’s EQ, dotted traces with no EQ :
The remaining differences are still audible. And in fact it’s debatable whether or not applying the preset made them sound closer to each others past a few kHz. In the 5-6.5kHz region it actually made them diverge more
(well if you normalise at 500Hz that is).
And remember that we’re talking here about two Sennheiser open back passive headphones with fairly low seatings variation and which already adhered quite well to the target.
With other headphones the differences when applying Oratory’s presets may be higher.
Another illustration of a typical problem. Apple decided to reinvent the yoke mechanism with the AirPods Max and this is only bringing a ton of problems to it as their solution applies all clamping force at the top of the earcup, and then tries to re-balance the pressure and seal the bottom by spring loading the cups. On my head with large temples and jaws but a narrow neck this results in the pads being a lot more compressed at the top front than at the rear bottom (but I'm still getting a good seal). This is how the APM varies between when I let them naturally sit on my head (green), and when I deliberately apply pressure at the bottom rear of the cups (orange), with both types of mic :
You notice however that the FR curve below 800hz or so is invariant. That’s because in that range Apple uses the inward facing mic to adjust the FR in real time to reach a precise target (what they call Adaptive EQ). Other ANC headphones do this as well (Bose, Sony).
But above that ? Significant variation (the APM isn’t the worst I have at home in that regard BTW). Someone else may get a result closer to one of these traces depending on their head shape.
Other than designing headphones better for low variability I’m not sure what would be the most effective solution. At lower frequencies headphones with an open front volume tend to be quite insensitive to seal variation, but not necessarily to pads compression. ANC headphones seem to have the best solution, using their inwards facing mic for real-time adjustments. Above that perhaps the same sort of sensors that could be used to gather anatomical data within the headphones’ earcup could be used for real time compensation as well ?
Anyway, if all of that is happening, you just can’t plug in a generic, baked in HRTF preset on a pair of headphones that may very well measure differently on your head than on a test rig and hope for the best. I have a feeling that we may really need both these problems tackled first to get truly effective sound virtualisation on headphones.