Apple AirPods Max Review (Noise Cancelling Headphone)

[Axo1989]

It's also, for the APM, largely related to HPTF, not HRTF.

In both the APM and APP2 cases they're facing HPTF / eardrum HRTF mismatch issues (and possibly undershooting - on average ! - a bit the response in the 1-5kHz range to avoid any excess), but the underlying mechanisms for that mismatch are entirely different.



I think that Sean Olive was rather talking about the issue of inter-individual, in situ FR variation, rather than anything related to active systems.

The APM's active systems, relying on the internal mic, seem to only operate up to around 1kHz, in all modes. In that band there shouldn't be a lot of reasons to have the FR vary from one individual to the other based on anatomical features, the goal rather being to maintain the response as consistent as possible regardless of coupling / leakage. Above 1kHz I haven't seen much evidence that they're doing anything with active systems.

The APP2's active systems operate up to around 4-5kHz in all modes and seem to try to ensure a consistent response at the eardrum for all individuals, which actually could very well be not desirable in the 1-5kHz region, but still better than letting the response vary in a way similar to passive IEMs or even worse active IEMs with a "classic" feedback system operating up to 500-800Hz (which is great for controlling leakage, but amplifies HPTF / eardrum HRTF mismatch in the mids in some cases).

Bose's CustomTune feature operates up to around 5-6kHz or so, and this time does seem to try to individualise the response at the eardrum in the 1-6kHz band, at least based on ear canal length, but it's a one-off measurement at startup, while Apple's systems are continuously updated using playback content as measurement signal.

Whether these active systems operating past 1kHz are doing a good job or not across a large sample of individuals remains to be tested outside of these companies R&D's labs .

In all three cases ear geometry is not used with their onboard active systems, but as @staticV3 mentioned, Apple uses the iPhone's depth sensor to at least partly individualise the HRTF map / ITD / ILD (?) when using their binaural renderer.

Apple has a patent to reduce HPTF issues based on scans of one's face, but I don't think that it's been applied in their product. They also seem to have floated the idea of using structured light sensors inside the ear cup of their over-ears to have a rough image of the user's ear, but again it seems like a far-fetched idea rather than a directly applicable patent, and mostly aimed at identifying left and right ears.



If Harman 2018 also means adding compression, maybe . And "maybe", as you're facing HPTF issues anyway, so difficult to know for certain.



Not a given depending on the actual exact model and the part of the spectrum under consideration in general. But for the APM, just like with most ANC over-ears indeed, it's very reliable in the range where their active systems operate (here up around 800Hz), not so much above.

Apparently AirPods Max gen 2 is imminent. Some updates indicated there but I haven't researched fully yet,

 

[Axo1989]

It isn’t a LIDAR but a dot pattern projector plus a camera.
Anyway there is customization of ear gain in the settings as mentioned in posts above by many others with pros and cons

Whether you call it LiDAR or not seems to depend whether you interpret the acronym to be "light detection and ranging" or "laser imaging, detection, and ranging".

The difference between eg infrared and laser is basically range, so when we are talking about scanning one's ears with a phone, the distinction isn't meaningful.

 

[MayaTlab]

Apparently AirPods Max gen 2 is imminent. Some updates indicated there but I haven't researched fully yet,

Hopefully they'll be able to make the response more predictable past 1kHz, whether by passive or active means.

 

[Mauro]

Whether you call it LiDAR or not seems to depend whether you interpret the acronym to be "light detection and ranging" or "laser imaging, detection, and ranging".

The difference between eg infrared and laser is basically range, so when we are talking about scanning one's ears with a phone, the distinction isn't meaningful.

you are right. it’s not particularly meaningful. but it’s anyway wrong. I worked on dot pattern projectors so I can tell the difference. As it is an engineering-oriented forum, maybe it’s interesting also for someone else.

 

[staticV3]

It isn’t a LIDAR but a dot pattern projector plus a camera.

Here's what the iPhone's FaceID system can capture:

Creative's three photos of face, left pinna, right pinna are a far cry for estimating HRTF.

 

[Mauro]

Here's what the iPhone's FaceID system can capture:

Creative's three photos of face, left pinna, right pinna are a far cry for estimating HRTF.
View attachment 380493

These technologies are rapidly evolving. Especially lasers and diffractive optics. there are big investments in engineering and production going on. We all need to wait and see what’s next..

 

[IPunchCholla]

you are right. it’s not particularly meaningful. but it’s anyway wrong. I worked on dot pattern projectors so I can tell the difference. As it is an engineering-oriented forum, maybe it’s interesting also for someone else.

It would be interesting for me, if how exactly it was wrong could be illuminated more. Searches show everyone calling what the iPhone does LiDAR and or LIDAR with those refer to light and LASER respectively, while the Wiki then confuses things by saying both use lasers.

 

[Mauro]

It would be interesting for me, if how exactly it was wrong could be illuminated more. Searches show everyone calling what the iPhone does LiDAR and or LIDAR with those refer to light and LASER respectively, while the Wiki then confuses things by saying both use lasers.

LiDARs are based on a light pulse of a laser and then a photodiode reads the time of arrival of the reflected light. You can have a scanning mirror as well to scan in multiple dimensions.
Dot projectors are based on a laser which has on top a lens that thanks to its properties can diffract the laser light and produce an array of points and project them onto a surface without scanning element. The camera then is taking a picture.
Here there is something:

Face ID - Wikipedia

en.m.wikipedia.org

 

[Sean Olive]

Every DSP enabled headphone and IEM should include selective 3kHz personalization controls like this. When most products omit that this puts APM ahead of the competition. I'm not an Apple fanboy by any means but giving credit when it's due, and hope the competition catches on, leaves arbitrary presets and frequency bands to the mists of time.

I might have agreed with you until I did this experiment where listeners were given a knob to adjust a filter centered at 3kHz (Q = 2) and could adjust it in 0.25 dB increments over a range of +6 to -10 dB. This was an IE headphone tuned to the B&K 5128 DF response after they made adjustments to the bass and treble.

Here is the box plot of those adjustments. The mean level is 0.11 dB after 432 adjustments were made by 36 listeners with 3 tracks of music. I don't see a lot of evidence that on average people want to adjust the ear canal gain much beyond the target.

 

[Sean Olive]

How do they do that? Airpod max has mic to get this information from reflection? Or a Camera? How do they even know ear canal geometries of the person wearing it?

Ω

How do they do that? Airpod max has mic to get this information from reflection? Or a Camera? How do they even know ear canal geometries of the person wearing it?

I was referring to the tuning of the headphone based on a measurement on a standard test fixture like GRAS45CA or B&K5128 -- the HPTF.

Not as a result of some dynamic active sensor that adjusts for the listener's ear canal acoustics

 

[markanini]

I might have agreed with you until I did this experiment where listeners were given a knob to adjust a filter centered at 3kHz (Q = 2) and could adjust it in 0.25 dB increments over a range of +6 to -10 dB. This was an IE headphone tuned to the B&K 5128 DF response after they made adjustments to the bass and treble.

Here is the box plot of those adjustments. The mean level is 0.11 dB after 432 adjustments were made by 36 listeners with 3 tracks of music. I don't see a lot of evidence that on average people want to adjust the ear canal gain much beyond the target.



View attachment 380580

Do you have any more details of the test protocol to share?

 

[MayaTlab]

I might have agreed with you until I did this experiment where listeners were given a knob to adjust a filter centered at 3kHz (Q = 2) and could adjust it in 0.25 dB increments over a range of +6 to -10 dB. This was an IE headphone tuned to the B&K 5128 DF response after they made adjustments to the bass and treble.

Here is the box plot of those adjustments. The mean level is 0.11 dB after 432 adjustments were made by 36 listeners with 3 tracks of music. I don't see a lot of evidence that on average people want to adjust the ear canal gain much beyond the target.



View attachment 380580

Thanks for sharing ! Have you had the opportunity to also test, with IEMs, shifting the frequency of the ear canal resonance (2-4kHz), not just the gain ?

 

[Sean Olive]

Do you have any more details of the test protocol to share?

I hope to publish an AES paper on these experiments in the next few months.. when I add more listeners. But essentially it involves using a method of adjustment where the DF target is adjusted in 3-bands

 

[Sean Olive]

Thanks for sharing ! Have you had the opportunity to also test, with IEMs, shifting the frequency of the ear canal resonance (2-4kHz), not just the gain ?

No, but Gaetan Lorho looked into that for the over-ear headphone and found the gain was the dominant perceptual factor not the frequency of the resonance.

E-library page - AES

AES E-Library ← Back to search

aes2.org

 

[MayaTlab]

No, but Gaetan Lorho looked into that for the over-ear headphone and found the gain was the dominant perceptual factor not the frequency of the resonance.

E-library page - AES

AES E-Library ← Back to search

aes2.org

Indeed, but I'm specifically wondering about IEMs, since as you already measured and shared, they don't vary at the eardrum, with ear canal length, in a way similar to either loudspeakers and over-ears.
Basically, a target for over-ears wouldn't need to see its ear canal resonance shifted given that the headphones will naturally do it "automatically" for each user (more or less, perhaps less for some closed back designs), but an IEM target might.

That article was quite an eye-opener for me, that simulation in particular :

Headphone-To-Ear Transfer Function Estimation Using Measured Acoustic Parameters

This paper proposes to use an optimal five-microphone array method to measure the headphone acoustic reflectance and equivalent sound sources needed in the estimation of headphone-to-ear transfer functions (HpTFs). The performance of this method is theoretically analyzed and experimentally...

www.mdpi.com

Here I'm plotting the difference in the graph above between the different ear canal lengths for both the article's IEM (typical for passive IEMs), solid traces, vs the DT880, dotted traces, plotting the 22mm and 32mm lengths over the 27mm one. I've added in light red and blue two quick and janky measurements of my own using canal extensions of a different lengths for a 711 coupler, compared to the default one, measured with uncoupled sound sources, to confirm that the trend observed in the simulation for the DT880 was similar in nature at least to these uncoupled sound sources (not exactly in degree as the difference in length isn't exactly the same to begin with).

 

[Mauro]

I might have agreed with you until I did this experiment where listeners were given a knob to adjust a filter centered at 3kHz (Q = 2) and could adjust it in 0.25 dB increments over a range of +6 to -10 dB. This was an IE headphone tuned to the B&K 5128 DF response after they made adjustments to the bass and treble.

Here is the box plot of those adjustments. The mean level is 0.11 dB after 432 adjustments were made by 36 listeners with 3 tracks of music. I don't see a lot of evidence that on average people want to adjust the ear canal gain much beyond the target.



View attachment 380580

Is this with a smooth response up in the higher frequency range? What if there was a peak at 6-10kHz? Would the sample of people increase anyway the 3kHz peak? My point is that Apple Airpods Max seem to have some peaks based on your measurements that might influence the result. What is your take?

 

[Sean Olive]

Is this with a smooth response up in the higher frequency range? What if there was a peak at 6-10kHz? Would the sample of people increase anyway the 3kHz peak? My point is that Apple Airpods Max seem to have some peaks based on your measurements that might influence the result. What is your take?
View attachment 380591

As I said the adjustment was made using the B&K 5128 DF response as a baseline so it’s relatively smooth at HF.

I cannot say if the lack of energy in the lower treble in AAP Max is related to HF response. You can check out their scientific publications or ask the person in charge of headphone r&d at Apple? Let me know if you get answer

 

[Mauro]

As I said the adjustment was made using the B&K 5128 DF response as a baseline so it’s relatively smooth at HF.

I cannot say if the lack of energy in the lower treble in AAP Max is related to HF response. You can check out their scientific publications or ask the person in charge of headphone r&d at Apple? Let me know if you get answer

Fair enough!

 

[markanini]

I hope to publish an AES paper on these experiments in the next few months.. when I add more listeners. But essentially it involves using a method of adjustment where the DF target is adjusted in 3-bands

Your last work evaluating proposed BK5128 targets aimed to answer a different question. But one can infer from those results that a range of varying 3Khz level preference exists, given that the two tied targets differ ~4dB at 3kHz. If my personal anecdote is worth anything, my preference for a lower 3Khz level comes with longer listening sessions, or higher listening volumes.

 

[Sean Olive]

Indeed, but I'm specifically wondering about IEMs, since as you already measured and shared, they don't vary at the eardrum, with ear canal length, in a way similar to either loudspeakers and over-ears.
Basically, a target for over-ears wouldn't need to see its ear canal resonance shifted given that the headphones will naturally do it "automatically" for each user (more or less, perhaps less for some closed back designs), but an IEM target might.

That article was quite an eye-opener for me, that simulation in particular :

Headphone-To-Ear Transfer Function Estimation Using Measured Acoustic Parameters

This paper proposes to use an optimal five-microphone array method to measure the headphone acoustic reflectance and equivalent sound sources needed in the estimation of headphone-to-ear transfer functions (HpTFs). The performance of this method is theoretically analyzed and experimentally...

www.mdpi.com

View attachment 380586
Here I'm plotting the difference in the graph above between the different ear canal lengths for both the article's IEM (typical for passive IEMs), solid traces, vs the DT880, dotted traces, plotting the 22mm and 32mm lengths over the 27mm one. I've added in light red and blue two quick and janky measurements of my own using canal extensions of a different lengths for a 711 coupler, compared to the default one, measured with uncoupled sound sources, to confirm that the trend observed in the simulation for the DT880 was similar in nature at least to these uncoupled sound sources (not exactly in degree as the difference in length isn't exactly the same to begin with).

View attachment 380587

There is no dispute that the natural ear canal resonances for an open ear shift when the ear is occluded with an IEM depending on the insertion depth, etc. The past year we've spent some time modeling earphones in different ear canals (based on MRI scans) in COMSOL and compared the results with the headphones measured in 3D printed canals and the models are quite accurate.

The main challenges or research questions are:

1. Should the IE headphone response at DRP replicate the open ear canal response at DRP ? Can you quantify benefits in terms of sound quality?
2. How do you model an individual's ear canal and predict the open ear canal resonance at DRP without using a probe mic or having access to an MRI? How much does the impedance of the ear drum vary between individuals to make it important to include?

The current version of the HARMAN IE target curve is already based on an open ear canal resonance at DRP based measurements of sound sources in the room using an B&K 5128 which is supposed to be an "average" human ear. So the question is how much better is the experience when personalized to your own canal?

Samsung presented a paper on question 2. at the recent AES conference in Madrid

I would be interested in your take on this paper as I have my own thoughts.

E-library page - AES

AES E-Library ← Back to search

aes2.org