Thoughts/discussion on The Headphone Show & other audio-related livestreams

[ADUHF]

Measured at the eardrum, it would measure the same.

But you'd have to simulate through a DSP all of the myriad of modifications that speakers in a room make on the source signal due to the room, the distance, and the physiognomy of the listener. There are a whole lot of variables in speaker sound that affect the response that you aren't taking into account.

For instance... A tweeter at 20 feet is entirely different than a tweeter at 20cm. You would have to calculate the amount of high end roll off that is created by traveling over a longer distance and incorporate that into your EQ curve. And that would be the simplest part of the simulation. Room reflections and timing differences are massive. I think your theory is like trying to figure out how to cut an apple to make it into an orange.

If one out of two is all I can get, I'll take that.

 

[bigshot]

I’ll just listen to my speakers!

 

[MayaTlab]

A good point, sander99. With over-ear headphones though, I think the variations in HRTF would mostly be head and torso-related, rather than ear-related. And how noticeable those differences might be from individual to individual (away from the original HATS rig where the measurements were made) is hard to say.

If you had the ability to measure both the neutral loudspeakers, and the headphones at the DRP for each individual though, then I think my theory about them measuring the same could still apply. This would not be practical for most people though.

Highly unlikely that even with over-ears the headphones transfer function is similar to the head related transfer function. If that were the case then the measurements made near the DRP in that study with an Etymotic ER7C with the tested over-ears would show constant a constant deviation at least between the over-ears that were measured. https://www.aes.org/e-lib/browse.cfm?elib=16877
That isn't the case.

While there may be some tendencies for each individual, we can see that the individual traces don't vary between each others in a truly constant fashion (the way it would be constant between them if you measured six different near-fields speakers for example). Not that I would presume as to why that is the case, whether it's position over the ear, pad compression / wear, ear shape, IDK.

 

[ADUHF]

Welcome to the discussion, Maya. And thank you for posting the link and images to the above study!... Some very interesting stuff there.

Here's the introduction in case anyone wants to read a little more what this was all about...

https://www.aes.org/images/e-lib/thumbnails/1/6/16877_full.png

I think my comment about the HRTFs and things "measuring the same" was probably not all that clear in the above post that you referenced. So I'll try to explain a little better what I meant by that... In a just a bit.

If I'm reading the above paper correctly though, then the study you're referencing above basically compared the in-ear measurements of a variety of different individuals to each other, and also to the in-ear measurements made on a dummy head, to see what differences and similarities there are between all of them. And this test was repeated using several different headphones.

It looks like they also used two different methods for the measurements on the dummy head. Though it's not entirely clear to me how that was done. It looks like one set of measurements was made from the mic in the dummy head's ear simulator though, which is represented by the light blue curves on the first set of graphs. And the other set of measurements was perhaps made with their own in-ear mic, inserted into the dummy's ear like they did for the other individuals. And this 2nd set of measurements appears to be represented by the red curves.

It's probably also worth noting that they did not perform any in-ear measurements for loudspeakers in this study. So only headphones. (Which was sort of the crux of my previous comment. But I'll get to more on that in a bit.)

If I'm interpreting the results of this study correctly, then it looks like there was a fair amount of variation in the in-ear measurements of the different individuals. And also some differences between the mean values of all those individuals (the black curve on the first set of graphs), and the measurements made on the dummy head (blue and red curves).

And the measurements for each headphone also appear fairly distinct from one another, indicating a good degree of variation in their sound quality and tonal balance.

The fact that you can see some fairly distinct differences in the measurements between each one of the headphones also probably indicates that there is at least some commonality between the way different people's ears measure on the same headphone. IOW, we all have ears that are at least somewhat similar in shape, that will measure somewhat similarly at the eardrum. If this were not the case, and the measurements of different individuals were less consistent, then it would probably be harder to separate the sound signatures of the different headphone from one another in a random study like this. Because the measurements would probably be more all over the map.

It would've been nice if the measurements made on the dummy head were a little closer to the mean or average value of the actual individuals in the study. It is well known that most HATS rigs currently in use are only accurate within in a certain limited frequency range though. And not as reliable in the treble, and sub-bass frequency ranges. However, the measurements made with the dummy head appear to fall mostly within or at least close to the standard deviation of the individual measurements. Which is probably a good thing. (And hopefully the new B&K 5128 head and torso rigs will be a little more accurate, particularly in the treble range.)

 

[bigshot]

You sure do grab on tight to your theory!

 

[MayaTlab]

The fact that you can see some fairly distinct differences in the measurements between each one of the headphones also indicates that there is at least some commonality between the way different people's ears measure on the same headphone. IOW, we all have ears that are at least somewhat similar in shape, that will measure somewhat similarly at the eardrum. If this were not the case, and the measurements of different individuals were less consistent, then it would probably be harder to separate the sound signatures of the different headphone from one another in a random study like this. Because the measurements would probably be more all over the map.

My understanding is that the problem is that the deviation between listeners is inconsistent between the over-ears that were tested.

If the study was about six different near field speakers, for example in front of the listeners, in an anechoic room, you'd expect variation for the same speaker between individuals (obviously), you'd expect variations between speakers for the same individual (obviously), but you would also expect no variation in how each listener's HRTF modulate all six speakers' response.
In other terms, you would observe that at, for example, 4000hz, listener's A HRTF results in a curve 4dB above listener B for all six speakers. Or that at 6317hz, listener's B HRTF results in a response 6.7dB lower than listener C for all six speakers.
In other words if you know how any speaker measures at listener A's DRP, and how listener A and B's HRTF deviate from each other, you'd be able to calculate how any speaker would measure at listener B's DRP (in the context of the aforementioned near-field situation).

This is not what we see in this study for headphones. The variation between individuals is not constant across all headphones tested, even just for the over-ears. That may be for reasons other than the listeners' ear shape (head size, neck width, pad compression, position over the ears, whatever, I have no idea).
In other words, even if you know how headphones X measure at listener A's DRP, since there seems to be no such thing as a constant headphones transfer function, you can't calculate how they'll measure at listener B's DRP.
In other words, if you measure that headphones X and Y differ by 4dB at 6000hz at listener A's DRP, you can't assume that they'll differ by the same amount at listener B's DRP.
My interpretation of it is that even over-ears will struggle to excite our ears in a way that speakers or natural sound sources would and that the headphone transfer function is unlikely to ever match the head related transfer function, at least with passive means only.

I don't feel that the magnitude of the problem is that significant in an age when we're still dealing with headphones which FR curve are so rubbish that they can only be justified by having been designed for aliens. But maybe sufficiently significant for the last mile in fidelity. I'm not convinced for example that truly convincing surround sound simulation can be achieved without tackling it.

I've started measuring headphones with in concha mics for the response below 1khz on my own head, and I'm actually starting to make my own DIY mics to try to get meaningful measurements above that. Including early attempts at making DIY tube microphones of the kind used in that study. I have no idea whether I'll get measurements that make sense, and I'm not qualified to make it anything that serious and rigorous, but I'm quite interested in comparing various means of measuring headphones on head (for example blocked ear canal measurements vs. tube mics near the eardrum), and perhaps headphones vs. speakers.

 

[bigshot]

Speakers at a distance are all hitting the ear and body basically the same way. But headphones don't necessarily emit sound the same way into the ear. For instance, the direction the sound is aimed into the ear canal is different on a planar than a dynamic. And some headphones sit differently on the head, positioning the transducer differently relative to the outer ear and ear canal. Headphones are just an entirely different animal than speakers. I recently got AirPods Max, which have microphones to automatically correct the EQ for different ways the ear cups might interface with the head. When you are strapping your transducer directly to your ears, the tolerances are very tight and response can vary a lot. A little shift one way or the other can make a huge difference. That's why headphone response measurements need to be taken with a grain of salt. What you see on the paper, isn't necessarily what you hear with your ears the way it is with speakers. And that isn't even considering all the timing differences a room adds to the sound.

 

[MayaTlab]

I recently got AirPods Max, which have microphones to automatically correct the EQ for different ways the ear cups might interface with the head.

Works in real time up to around 1khz :
https://twitter.com/oratory1990/status/1343323936801644544?s=20
Similarly to some other ANC headphones :
https://diyaudioheaven.wordpress.com/headphones/measurements/bose/qc35-ii/

The AKG N90Q tried to use two feedback microphones per cup (only one for most ANC headphones) to attempt a non real time individualisation of the FR curve past 1khz with test tones : https://www.stereophile.com/content/akg-n90q-noise-canceling-auto-calibrating-over-ear-headphones
Since that system isn't in real time by definition it can't correct for positional variation during the same listening session, for example.
I don't have the chops to understand this paper (free access) well but it appears that it's an attempt to use only one feedback mic such as the ones already present on ANC headphones for limited, cautious customisation of the FR curve above 1khz - at least for the ear canal gain region : https://www.aes.org/e-lib/browse.cfm?elib=20605

 

[bigshot]

The AirPods Max has nine microphones. I'm not sure how many of those are used in adaptive EQ and which ones are intended for other purposes like voice though. It really is interesting to move them around on your head while they're playing. The response stays perfectly focused, even with quite radical position shifts. The other thing about these cans that is amazing is the transparency mode. It sounds totally natural.

 

[MayaTlab]

I'm not sure how many of those are used in adaptive EQ

The only one inside the earcup (gold part on the grill). https://guide-images.cdn.ifixit.com/igi/lDAAhbWSGNyJ2T3W.huge

The response stays perfectly focused, even with quite radical position shifts.

My opinion is that they behave like other headphones in terms of positional variation at higher frequencies. They're unlikely to have any way to compensate for that.
Besides Oratory's multiple reseats there's also Rtings : https://www.rtings.com/headphones/1-4/graph#16092/4007

 

[bigshot]

The low end is what is affected most by seal issues. And above 10-12kHz, it likely doesn't make much of a difference anyway. So there's only a few octaves there that aren't covered. In practice it does a fantastic job.

 

[Blackwoof]

Another question which often comes up when discussing the different compensation curves for headphones is-- why is a diffuse field curve so bright or brightly tilted in the treble and upper mids compared to the frequency responses of most of the better or more "neutral-souding" headphones? This is also something that Resolve didn't really attempt to explain in the above video. (And it's really asking the wrong question imho.) The answer though is that the diffuse field curve is simply the head, torso and ear's response to sound coming equally from all directions... And that's pretty much it. So nothing else.

Too many people who do reviews never point that out our ears colour the sound anyway hence why the Etymotic ER4 has that 10db bump at 2.7KHz. It same story with ER4's bass it flat because that how our eardrum would receive low freq sounds, It hard to engage when people turn that bit into a Dynamic vs BA war. By making claims that the Etymotic ER2SE warmer when it has the same tuning as the ER4S. If anything the BA's used in the ER4/ER3 have much better decay time which helps with fast complex sounds. While the ER2SE can sound muddy since It can't keep up with sudden bass notes in fast death metal or Hard EDM.

 

[redrol]

IEMs may work somewhat differently than over-ears on some of this. Because they have to make up a bit more for some of the missing effects of the outer ear than most over-ear headphones would.

I think its more than 'a bit'. 10db gain at say 2-3k is enormous gain (ear cup)

 

[bigshot]

IEMs are radically different than over ear headphones. And over ear headphones are even more radically different than speakers in a room. You really have to talk about each kind separately and optimize the curve specifically for the benefits and drawbacks of that particular kind of transducer. I don't think it's helpful to try to turn an apple into an orange. Instead, you should try to make that the best apple it can be.

 

[ADUHF]

Thank you for all the replies on this topic folks.

My understanding is that the problem is that the deviation between listeners is inconsistent between the over-ears that were tested.

If the study was about six different near field speakers, for example in front of the listeners, in an anechoic room, you'd expect variation for the same speaker between individuals (obviously), you'd expect variations between speakers for the same individual (obviously), but you would also expect no variation in how each listener's HRTF modulate all six speakers' response.
In other terms, you would observe that at, for example, 4000hz, listener's A HRTF results in a curve 4dB above listener B for all six speakers. Or that at 6317hz, listener's B HRTF results in a response 6.7dB lower than listener C for all six speakers.
In other words if you know how any speaker measures at listener A's DRP, and how listener A and B's HRTF deviate from each other, you'd be able to calculate how any speaker would measure at listener B's DRP (in the context of the aforementioned near-field situation).

This is not what we see in this study for headphones. The variation between individuals is not constant across all headphones tested, even just for the over-ears. That may be for reasons other than the listeners' ear shape (head size, neck width, pad compression, position over the ears, whatever, I have no idea).
In other words, even if you know how headphones X measure at listener A's DRP, since there seems to be no such thing as a constant headphones transfer function, you can't calculate how they'll measure at listener B's DRP.
In other words, if you measure that headphones X and Y differ by 4dB at 6000hz at listener A's DRP, you can't assume that they'll differ by the same amount at listener B's DRP.
My interpretation of it is that even over-ears will struggle to excite our ears in a way that speakers or natural sound sources would and that the headphone transfer function is unlikely to ever match the head related transfer function, at least with passive means only.

I don't feel that the magnitude of the problem is that significant in an age when we're still dealing with headphones which FR curve are so rubbish that they can only be justified by having been designed for aliens. But maybe sufficiently significant for the last mile in fidelity. I'm not convinced for example that truly convincing surround sound simulation can be achieved without tackling it.

I've started measuring headphones with in concha mics for the response below 1khz on my own head, and I'm actually starting to make my own DIY mics to try to get meaningful measurements above that. Including early attempts at making DIY tube microphones of the kind used in that study. I have no idea whether I'll get measurements that make sense, and I'm not qualified to make it anything that serious and rigorous, but I'm quite interested in comparing various means of measuring headphones on head (for example blocked ear canal measurements vs. tube mics near the eardrum), and perhaps headphones vs. speakers.

And thank you also for explaining this a bit more, Maya. I think I understand what you're driving at now about the differences in the way speakers measure vs. headphones.

As you and others have pointed out, there are some variables with headphones that don't exist with speakers. Such as the fit, seal, angle of the drivers and so forth, which could effect how consistently they measure from individual to individual. And also from one usage to another, by the same individual. As bigshot and some others like to point out though, there are also variables with speakers, which do not exist with headphones. For example, if you measure them in one room, you'll generally get a different result if you measure them in another room. Or even from a different location in the same room.

I think it's good to be aware of all these potential pitfalls and variables, and how they could impact the final result. So you can minimize them or their effects as much as possible. And (fwiw) I don't really expect to ever achieve an exact reproduction of what I hear on a pair of loudspeakers in a pair of headphones. As bigshot likes to repeatedly point out, that is a difficult, if not impossible task. And it's something I'd rather leave up to all the programming wizards out there, to figure out instead... So that isn't really my objective here.

My only objective is to try to get as close as possible to matching or approximating the tonal balance of a neutral speaker in a room, within the obvious limitations of all these different variables.

I think the Harman curve was a step forward in some ways on this. Because it adds a little bit more of that darker tilt that you get when you put a speaker in a room. I think it overdoes it in some ways though, esp. maybe in the treble. And maybe also underdoes it (is that even a word?) in some other ways. So that's why my quest for more info on this subject, and a better model for this type of sound continues. And I'm also hoping one of the side benefits of that will be a mitigation of the impact of some of the kinds of variables that you both are bringing up, which are likely to occur further downstream in this process.

It is possible that this is all just foolishness, and a complete waste of time though. Stranger things have certainly happened.