Is it possible to measure HRTF by photo?

[Volutrik]

Hi. I recently tried out Immerse Virtual Studio and they ask you to take a photo of your ear to generate an HRTF using AI processing. The results were pretty amazing! It really does sound like I'm in that room. But I was wondering if there's an option to do this just for making the headphones' frequency response matched to my personal HRTF, instead of simulating a room, where you take a photo and it generates your HRTF. If not, does anyone know if I can get my HRTF from the Immerse Virtual Studio's servers too? I've heard about Impulcifier, but it requires an in-ear microphone and monitors. Since my monitors aren't even good, that wouldn't be an option. I just wanted to EQ to my HRTF

 

[IAtaman]

I would imagine that would probably be called ear related transfer function.

Apple has a feature with which you can take a photo of your head and your ears for personalized spatial audio. With high speed real time processing gettting smaller and cheaper, I think there is a lot of good reason to believe personalized audio to be the future.

Listen with Personalized Spatial Audio for AirPods and Beats - Apple Support

With Personalized Spatial Audio, you can use the TrueDepth camera on your iPhone to create a personal profile for Spatial Audio that delivers a listening experience tuned just for you.

support.apple.com

 

[thewas]

Creative had done such attempts 5 years ago https://mobilesyrup.com/2018/01/12/creative-labs-holography-technology/
and Genelec also offers currently such https://www.genelec.com/aural-id

 

[MayaTlab]

I've heard about Impulcifier, but it requires an in-ear microphone and monitors. Since my monitors aren't even good, that wouldn't be an option.

Impulcifier creates a binaural room impulse response, which is conceptually a bit different from mapping your own HRTF. I am not certain that you can easily measure your own HRTF in a typical room.

I just wanted to EQ to my HRTF

Strictly speaking HRTF is a map of how FR is affected by direction, so when you say that you want to EQ to your HRTF, I'm not certain that it makes total sense. What I guess you're trying to do is to EQ headphones so that they deliver a FR at your eardrum that is aligned with how your own anatomical features would affect what would be a desirable FR for you.

Some people have advanced that a Diffuse field HRTF + preferential tilt and / or shelves are a good way to capture an individual's anatomical features and derive from it a desirable curve for that individual, but I don't think that it's been fully vetted by listening tests vs. other approaches to derive an individualised target (such as measuring the in-room response of a decent pair of speakers in a decent listening room at your own eardrum as a baseline + preferential tilts and / or shelves). But capturing your own DF HRTF might be quite difficult as well .

One thing that I've toyed with is to measure different signals at different degrees of azimuth and elevation after feeding them through Logic Pro's several Dolby Atmos binaural renderers (crucially including Apple's personalised one), as this is an effective way to repeatedly locate a sound source in space with numerical entries, but this will not allow you to "measure" one's HRTF as these renderers aren't designed to create anechoic data to begin with.

 

[iGude]

Think about this:
They want you to take a photo of your ear and don’t require any size reference, i.e. they have no information about the dimensions of your ear. At the same time, these dimensions define your individual HRTF to a large extent…
They only need the photo of one ear. At the same time, a large amount of people have significant differences between the left ear HRTF and the right ear HRTF…

…marketing BS.

 

[MayaTlab]

Think about this:
They want you to take a photo of your ear and don’t require any size reference, i.e. they have no information about the dimensions of your ear. At the same time, these dimensions define your individual HRTF to a large extent…
They only need the photo of one ear. At the same time, a large amount of people have significant differences between the left ear HRTF and the right ear HRTF…

…marketing BS.

I don't know how Embody acquires the data but there is some variation in the way these systems capture data. Genelec's is video based unless I'm mistaken, for example, and Apple's uses the depth camera from the FaceID system (which also means that it's mostly immune to light level differences).

Also, if they're machine learning based, from what I understand they only need to find strong enough correlations between some anatomical variables and measured HRTFs. ie if "neck width" just happens to correlate well with some aspects of HRTF variation that are in reality caused by other anatomical variables, it still works. Which could be convenient as I doubt that any of these different ways to capture data are adequate enough to get a fine-grained 3D mesh of one's pinna / concha.

 

[iGude]

I don't know how Embody acquires the data but there is some variation in the way these systems capture data. Genelec's is video based unless I'm mistaken, for example, and Apple's uses the depth camera from the FaceID system (which also means that it's mostly immune to light level differences).

Also, if they're machine learning based, from what I understand they only need to find strong enough correlations between some anatomical variables and measured HRTFs. ie if "neck width" just happens to correlate well with some aspects of HRTF variation that are in reality caused by other anatomical variables, it still works. Which could be convenient as I doubt that any of these different ways to capture data are adequate enough to get a fine-grained 3D mesh of one's pinna / concha.

Correct, other solutions based on LiDAR or 3D mesh reconstruction from videos do have information about the dimensions of both ears.

My statement was referring to the product mentioned in the OP and other solutions based on simple photos without size reference. Those can safely be considered snake oil.

BTW I read about somebody who tested such a photo based solution, one time with a photo of his own ear and one time with a photo of the ear of his dog (or cat, don’t remember). Neither did the software complain, nor could he hear any difference between those two ‘ears’ …

 

[Volutrik]

Some people have advanced that a Diffuse field HRTF + preferential tilt and / or shelves are a good way to capture an individual's anatomical features and derive from it a desirable curve for that individual, but I don't think that it's been fully vetted by listening tests vs. other approaches to derive an individualised target (such as measuring the in-room response of a decent pair of speakers in a decent listening room at your own eardrum as a baseline + preferential tilts and / or shelves). But capturing your own DF HRTF might be quite difficult as well .

Hey! That's really interesting. I matched my headphones' frequency response to the Diffuse Field target available at the Auto EQ Github repository and tested it against the Harman target, using Waves NX plugins. I actually think DF is better at reproducing the sense of space, it sounds more open while the Harman target makes it sound muffled, closed. But then I searched a bit more and found Oratory1990's answear about a question asking if Diffuse Field is accurate:

"The diffuse-field curve certainly isn't.
It's a reference, but it's not a target.
As in: knowing the diffuse-field transfer function of a particular measurement setup is generally useful, but this does not mean that headphones should be tuned to the diffuse-field curve (e.g. should produce a linear frequency response when measured with a diffuse-field compensated measurement setup).

Tuning a headphone to the diffuse-field curve is only necessary for very specific applications, e.g. when listening to recordings made with a diffuse-field equalized binaural microphone such as the Neumann KU81 or the Neumann KU100."

Even though I preferred Diffuse Field over the Harman target, it got me wondering about this, because Waves decided to use the Harman target as a standard for the built-in headphone corrections.

For binaural/room simulation stuff, is it better to use the Harman target or DF? Oh! Also, is that Diffuse field HRTF you talked about different from the normal one? (This one https://raw.githubusercontent.com/jaakkopasanen/AutoEq/master/compensation/diffuse_field.png)
Maybe it would make sense to use the Free Field compensation to use with such a plugin, because in Free Field, the sound is coming directly to the microphone without any reflection interference and the plugin is simulating the reflections of the room?
I'm all ears! Really curious to learn

 

[markanini]

Hey! That's really interesting. I matched my headphones' frequency response to the Diffuse Field target available at the Auto EQ Github repository and tested it against the Harman target, using Waves NX plugins. I actually think DF is better at reproducing the sense of space, it sounds more open while the Harman target makes it sound muffled, closed. But then I searched a bit more and found Oratory1990's answear about a question asking if Diffuse Field is accurate:

"The diffuse-field curve certainly isn't.
It's a reference, but it's not a target.
As in: knowing the diffuse-field transfer function of a particular measurement setup is generally useful, but this does not mean that headphones should be tuned to the diffuse-field curve (e.g. should produce a linear frequency response when measured with a diffuse-field compensated measurement setup).

Tuning a headphone to the diffuse-field curve is only necessary for very specific applications, e.g. when listening to recordings made with a diffuse-field equalized binaural microphone such as the Neumann KU81 or the Neumann KU100."

Even though I preferred Diffuse Field over the Harman target, it got me wondering about this, because Waves decided to use the Harman target as a standard for the built-in headphone corrections.

For binaural/room simulation stuff, is it better to use the Harman target or DF?

I haven't seen recordings that require something different from stereophonic setup for playback, maybe I don't know where to look fot this, but it seems rare. It's pretty much a universal effort for audio engineers to make content that can sound good everywhere, home stereos, headphones, venues. A concept going against that ethos is strange to me, but might be part of some companies marketing to sell a customers on an exclusive ecosystem.

 

[MayaTlab]

Hey! That's really interesting. I matched my headphones' frequency response to the Diffuse Field target available at the Auto EQ Github repository and tested it against the Harman target, using Waves NX plugins. I actually think DF is better at reproducing the sense of space, it sounds more open while the Harman target makes it sound muffled, closed. But then I searched a bit more and found Oratory1990's answear about a question asking if Diffuse Field is accurate:

"The diffuse-field curve certainly isn't.
It's a reference, but it's not a target.
As in: knowing the diffuse-field transfer function of a particular measurement setup is generally useful, but this does not mean that headphones should be tuned to the diffuse-field curve (e.g. should produce a linear frequency response when measured with a diffuse-field compensated measurement setup).

Tuning a headphone to the diffuse-field curve is only necessary for very specific applications, e.g. when listening to recordings made with a diffuse-field equalized binaural microphone such as the Neumann KU81 or the Neumann KU100."

Even though I preferred Diffuse Field over the Harman target, it got me wondering about this, because Waves decided to use the Harman target as a standard for the built-in headphone corrections.

For binaural/room simulation stuff, is it better to use the Harman target or DF? Oh! Also, is that Diffuse field HRTF you talked about different from the normal one? (This one https://raw.githubusercontent.com/jaakkopasanen/AutoEq/master/compensation/diffuse_field.png)
Maybe it would make sense to use the Free Field compensation to use with such a plugin, because in Free Field, the sound is coming directly to the microphone without any reflection interference and the plugin is simulating the reflections of the room?
I'm all ears! Really curious to learn

More recent suggestions to base targets on the DF HRTF response are not simply using the DF HRTF of a mannequin as a target, which indeed has already largely been demonstrated as being disliked by most people. They also introduce preferential tilts and / or shelves, just as Harman did after equalising headphones to a target that corresponds to the flat in-room response at a mannequin's DRP. You can have an idea of the reasoning behind it here (Mad_Economist posts) : https://audiosciencereview.com/foru...target-you-can-try-the-eqs.43209/post-1537951

It has not been vetted by listening tests, but nothing should prevent you from using that approach instead of Harman's work and make up you own mind what you prefer, or even introducing reasonable deviations from both of these approaches.

In all cases since I'm not certain that it will be easy to derive your own DF HRTF from the current visual based approaches that we've seen, either because the data capture isn't good enough or because it's not been made available to the end user, and since you currently aren't in possession of a speaker system that you think is good enough to "do a Harman", then it could be a bit difficult to try to use your own anatomy as a basis for a personalised target.