• WANTED: Happy members who like to discuss audio and other topics related to our interest. Desire to learn and share knowledge of science required. There are many reviews of audio hardware and expert members to help answer your questions. Click here to have your audio equipment measured for free!

Master Complaint Thread About Headphone Measurements

adude995

Member
Joined
May 17, 2019
Messages
15
Likes
5
Location
aut + fin
Yes, and even for different types of phones.
The benefit of measuring at the eardrum is, that I don't care where the sound comes from.
An over-ear with the pinna and ear-canal in the signal path, or just an in ear that plays directly at the eardrum, if it sounds the same, I perceive it as correct.

As with target vs correction curves, you are right. Both definitions are used the same way.
Although I would say that in our case, one is the invers of the other.
I don't think the harman bass is not linear, but the missing crosstalk especially in the lows, we have to correct. (I know there is an own thread about it)
 

solderdude

Grand Contributor
Joined
Jul 21, 2018
Messages
15,891
Likes
35,912
Location
The Neitherlands
The benefit of measuring at the eardrum is, that I don't care where the sound comes from.

Of course, but it is only accurate with that specific measurement system. I know research has been done to make it as 'average' as possible but it is obviously wrong. Measure the same headphone on various HATS, rigs with pinna + ear canal and use their respective DF corrections for instance and the corrected as well as the eardrum level results differ. So it is only correct to a standard one hopes is accurate. For headphones this is a much taller order than for speakers measured using the same measurement system.

don't think the harman bass is not linear, but the missing crosstalk especially in the lows, we have to correct.

If that were the case one would not expect the bass boost to be as it is because if it crosstalk it would not be as low as it is but there would be a gentle lift already occurring at 2kHz and not at 200Hz.
That lift clearly is preference based. Could be emulating bass boost in a room, could be lack of tactile feel, could be a combination but don't see why it would be crosstalk.
 
D

Deleted member 16543

Guest
I said that, because if I now make a measurement on my ear fixture, it will definitly look different than on another fixture.
And since we don't want to measure different silicon ears, but the headphones, we need to take the ear part out of the measurement, which i don't know and is not easy to identify in generel.

Exactly, and because the impulse response is a compound of an amplitude and it's phase (or the other way round) either the impulse response or the phase would be nice to know.
I don't think inaccurate is the correct term. A measurement is a measurement. It exists and is validated in and of itself (if it was made correctly).
What you are referring to is a consequence of the fact that this is a field of research that is still short of some big findings to become fully appreciated and useful.
For example, your rig (the one with the ear canals) can still tell you if headphones A are brighter or darker than headphones B.
What it can't tell you yet is how close it comes to what a balanced sounding pressure wave should measure like at the eardrum. Because it's missing a target curve.
A target curve to me is exactly that. The frequency response at the eardrum of a sound pressure that is considered balanced sounding. I'll try to avoid another spin-off conversation about speakers, but alas that's where the most reliable data as far as balanced sound comes from, to this day.
So the target curve to me has already, intrinsically and undistinguishably, the acoustic properties of the measurement rig embedded in itself.
The correction would be the EQ necessary (both amplitude and phase) to make a certain headphone measurement match with the target curve.

I understand the want for a measurement that is just for the headphone, 'cleaned' from the pinna and canal contribution, but that would be of little value in terms of translation to a sonic meaning to the listener. Because the behavior of a sound source in a certain setting (how headphones sound when worn on our head) can only be measured meaningfully when they are in fact placed and measured on a (mannequin) head.
Just consider the difference in the behavior of the diaphragm when it's used in free air or in a more or less enclosed space such as it's the case when you wear headphones. These differences can't be corrected for by a lump factor that represents only the contribution of the rig. The headphone behavior is rig specific and can't be measured in and of itself (which is also the reason why we need a rig that accurately represents a human head to make the most accurate measurements possible).

If you do research at a university and your department has enough funds, I'd suggest getting a B&K 5128 and try to find a target for it. This involves the use of speakers that have a balanced response. If you don't know what I'm talking about we can chat privately. I can most definitely land a hand here.
You can also use your own rig and save the $41K, provided that it's a rig that has accurate ear canals and made of material that bends similarly to a real pinna (I could help here too, but only on the first requirement, unfortunately).
 

Robbo99999

Master Contributor
Forum Donor
Joined
Jan 23, 2020
Messages
6,875
Likes
6,673
Location
UK
If you do research at a university and your department has enough funds, I'd suggest getting a B&K 5128 and try to find a target for it. This involves the use of speakers that have a balanced response.
I like that idea! Get the uni folks to do it, perfect. Then there has to be a person/source who is using B&K 5128 to measure headphones so that the community can use the EQ's......quite a few steps to complete and time involved, I'm not gonna hold my breath, maybe in a few years if we're lucky this might happen & have had a significant number of headphones in a database. I don't think there's a huge amount of value in the B&K 5128 though because it's selling point is more accurate measurements above 10kHz, but that's exactly the area where headphone placement and person to person variation differs the most.....therefore any "accuracy" in this region is by definition lost. GRAS system is fine (Amir/Oratory/Crinacle/Resolve/etc).
 
D

Deleted member 16543

Guest
The
I like that idea! Get the uni folks to do it, perfect. Then there has to be a person/source who is using B&K 5128 to measure headphones so that the community can use the EQ's......quite a few steps to complete and time involved, I'm not gonna hold my breath, maybe in a few years if we're lucky this might happen & have had a significant number of headphones in a database. I don't think there's a huge amount of value in the B&K 5128 though because it's selling point is more accurate measurements above 10kHz, but that's exactly the area where headphone placement and person to person variation differs the most.....therefore any "accuracy" in this region is by definition lost. GRAS system is fine (Amir/Oratory/Crinacle/Resolve/etc).

The lack of ear canal affects way below 10 kHz (in a direction dependant way), so agree to disagree.
 

Robbo99999

Master Contributor
Forum Donor
Joined
Jan 23, 2020
Messages
6,875
Likes
6,673
Location
UK
The

The lack of ear canal affects way below 10 kHz (in a direction dependant way), so agree to disagree.
I'm pretty sure GRAS has an ear canal.
 
D

Deleted member 16543

Guest
I'm pretty sure GRAS has an ear canal.

Somewhat. The entrance of the ear canal is anthropomorphic, but about 1cm in the canal is connected to a more traditional, cylindrical section coupler.

Also, it lacks the head shadow effect that would be necessary to figure out the target curve starting from balanced speakers.
So I really think the 5128, with all its lack of target curve and reported difficulty in measuring IEMs, is where it's at, at the moment.
 

Robbo99999

Master Contributor
Forum Donor
Joined
Jan 23, 2020
Messages
6,875
Likes
6,673
Location
UK
Somewhat. The entrance of the ear canal is anthropomorphic, but about 1cm in the canal is connected to a more traditional, cylindrical section coupler.

Also, it lacks the head shadow effect that would be necessary to figure out the target curve starting from balanced speakers.
So I really think the 5128, with all its lack of target curve and reported difficulty in measuring IEMs, is where it's at, at the moment.
Well I'd certainly like to compare EQ's done on 5128 using a well determined Target Curve along the lines of the Harman process and then compare those vs the same done on GRAS, but really you'd need someone to measure the same unit (not model) of headphone on both GRAS and 5128 and then you'd try out the EQ of both using the same model of headphone. I'm not convinced 5128 is gonna be better, I'd like to try out the EQ's, but I doubt this work is really gonna be done, and not in the way I outlined.

GRAS do have versions where it has the complete dummy head, have a look at the bottom of Oratory's pdf file here, you can see it says GRAS45BC KEMAR which is the version with the dummy head. https://www.dropbox.com/s/033kxyf8a0o3iv6/AKG K702.pdf?dl=0

5128 is of no use until there's a Target Curve done for it with the same validity as the Harman research, and also no use until someone starts measuring headphones and builds up a significant database of measurements that is free in the public domain - probably also with provided EQ's.....so GRAS is where it's at right now, 5128 may become a thing in the future, but you're talking years and I think we'll be in the personalised HRTF domain by then using DSP albeit that might need to rely on headphone measurements on ear simulators as part of the base still.
 
Last edited:

ADU

Major Contributor
Joined
Jul 29, 2021
Messages
1,587
Likes
1,086
Well I'd certainly like to compare EQ's done on 5128 using a well determined Target Curve along the lines of the Harman process and then compare those vs the same done on GRAS, but really you'd need someone to measure the same unit (not model) of headphone on both GRAS and 5128 and then you'd try out the EQ of both using the same model of headphone. I'm not convinced 5128 is gonna be better, I'd like to try out the EQ's, but I doubt this work is really gonna be done, and not in the way I outlined.

GRAS do have versions where it has the complete dummy head, have a look at the bottom of Oratory's pdf file here, you can see it says GRAS45BC KEMAR which is the version with the dummy head. https://www.dropbox.com/s/033kxyf8a0o3iv6/AKG K702.pdf?dl=0

5128 is of no use until there's a Target Curve done for it with the same validity as the Harman research, and also no use until someone starts measuring headphones and builds up a significant database of measurements that is free in the public domain - probably also with provided EQ's.....so GRAS is where it's at right now, 5128 may become a thing in the future, but you're talking years and I think we'll be in the personalised HRTF domain by then using DSP albeit that might need to rely on headphone measurements on ear simulators as part of the base still.

This is beginning to get a bit off topic for this thread, Robbo99999. Which I know is something you don't like. But if you believe that the current Harman curve is accurate, then you should be able to take Amir's, or Oratory's, or Jaakko's, or someone else's Harman-based EQ corrections (whichever datasets you think are the most reliable), and then simply apply their recommended EQ settings to the 5128 measurements of the same headphones. And then compute the average frequency response of those 5128 headphone measurements with the EQ corrections applied to derive a similar sounding target response curve for the 5128 rig.

I believe that the Harman curve is not completely correct though. And can use some improvement in some spots, especially in the higher frequencies (as I've said on several occasions here). And believe that an approach like the one above would carry over those potential errors or inaccuracies to the new 5128 data, possibly making them even worse. So I prefer to use a different approach which is based more on Harman's process of estimating the sound power responses of loudspeakers (known as "spinorama"). And using that as my target for the 5128 diffuse field compensated measurements instead.

It may not be 100% accurate either. But at least it will not carry over any potential errors or inaccuracies in the current Harman curve. And would probably be quite a bit more accurate than the approach above (using the Harman EQ curves). So there's really nothing to be lost by using this other approach... It can't work any worse than using the current Harman curve as a model. And has the potential to be quite a bit more accurate than the Harman curve, particularly in the higher frequencies. So I don't really see where there's any downside to it. (?)

If you believe the current Harman curve is correct though, then the first approach discussed above of simply applying the current Harman EQ curves to the 5128 measurements should work just fine for deriving your 5128 target response curve. In addition to Amir's 5128 measurements, there are also some other measurements made by Jude for the 5128 rig in some of his recent headphone reviews, and also a couple more in his Head-Fi Audio Measurements Lab that you could use to begin this process.

Using sound power is a more accurate way to go though, imho. With fewer potential downsides, and alot more potential upside!
 
Last edited:

Robbo99999

Master Contributor
Forum Donor
Joined
Jan 23, 2020
Messages
6,875
Likes
6,673
Location
UK
This is beginning to get a bit off topic for this thread, Robbo99999. Which I know is something you don't like. But if you believe that the current Harman curve is accurate, then you should be able to take Amir's, or Oratory's, or Jaako's, or someone else's Harman-based EQ corrections (whichever datasets you think are the most reliable), and then simply apply their recommended EQ settings to the 5128 measurements of the same headphones. And then compute the average frequency response of those 5128 headphone measurements with the EQ corrections applied to derive a similar sounding target response curve for the 5128 rig.
This is what Jaakopasenen did to determine his "Harman Curves" for say Innerfidelity measurements. I tried it ages ago on my K702 and it's "OK", but I've sent two K702's to Oratory to measure on his GRAS and the EQ corrections are different - so it's not a totally valid method of converting target curves from one measurement rig to another.....I think Oratory described the phenomenon as "acoustic impedance" was different between rigs and therefore that was the reason that you couldn't accurately do this type of conversion between rigs. So the method you propose is better than a kick in the head if your headphone has not been measured on a GRAS, but still better to use proper Harman Curves with the proper measurement rig - ie GRAS.
 

ADU

Major Contributor
Joined
Jul 29, 2021
Messages
1,587
Likes
1,086
This is what Jaakopasenen did to determine his "Harman Curves" for say Innerfidelity measurements. I tried it ages ago on my K702 and it's "OK", but I've sent two K702's to Oratory to measure on his GRAS and the EQ corrections are different - so it's not a totally valid method of converting target curves from one measurement rig to another.....I think Oratory described the phenomenon as "acoustic impedance" was different between rigs and therefore that was the reason that you couldn't accurately do this type of conversion between rigs. So the method you propose is better than a kick in the head if your headphone has not been measured on a GRAS, but still better to use proper Harman Curves with the proper measurement rig - ie GRAS.

This is why you'd compute an average for the responses of multiple headphones, with the Harman EQ curves applied.

If you believe that the Harman target and GRAS EQ curves derived from it are correct, then you should be able to reverse engineer a similar sounding target response curve for another type of measurement rig (like the 5128), by simply applying those Harman/GRAS-based EQ curves to the headphone data on the other rig. And then computing the average response of those EQ-corrected headphone measurements for the new rig.

The more headphone measurements you have for doing this, the more reliable this type of translation would probably be. The theory is totally sound though... At least for over-ear type headphones. (IEMs might possibly be a little less reliable, due to variations in the way different rigs simulate the shape and response of the ear canal.)

If you don't believe the Harman curve, and EQ curves derived from it are completely accurate though, then an approach like this would probably be mostly a waste of time and energy, and good plotting skills.

This is just my opinion, but I think sound power probably offers a much easier, and potentially more accurate route to a neutral (DF) response curve for the 5128 than the above. YMMV though on this.
 
Last edited:

MayaTlab

Addicted to Fun and Learning
Joined
Aug 15, 2020
Messages
924
Likes
1,512
In regards to Harman's research and the 5128, I believe that Sean Olive may have a word on the subject during the following conference :
 
  • Like
Reactions: ADU

Robbo99999

Master Contributor
Forum Donor
Joined
Jan 23, 2020
Messages
6,875
Likes
6,673
Location
UK
This is why you'd compute an average for the responses of multiple headphones, with the Harman EQ curves applied.

If you believe that the Harman target and GRAS EQ curves derived from it are correct, then you should be able to reverse engineer a similar sounding target response curve for another type of measurement rig (like the 5128), by simply applying those Harman/GRAS-based EQ curves to the headphone data on the other rig. And then computing the average response of those EQ-corrected headphone measurements for the new rig.

The more headphone measurements you have for doing this, the more reliable this type of translation would probably be. The theory is totally sound though.

If you don't believe the Harman curve, and EQ curves derived from it are completely accurate though, then an approach like this would probably be mostly a waste of time and energy, and good plotting skills.
As I mentioned in my last post, Jaakopasenen already did this, but instead by comparing headphone measurements of many different models vs the same measurements of said models on another rig, this is how he came up with this "Harman Curves" for the various measurement rigs as a means of converting from GRAS. If you need to find out more, speak to Jaakopasenen about it. I don't have any more to add on your ideas of converting between measurement rigs, which I know you had enthusiasm for comparing Diffuse Field measurements.....I don't have anything to add on anything more in relation to these topics except - use GRAS measurements and the Harman Curve (Oratory / Amir / Crinacle / Resolve) and if your headphone has not been measured yet send it to Oratory to be measured, and then if all else fails use Jaakopasenen's site for his "Harman Curves" that are based on Innerfidelity measurements (ie not GRAS). I know you're like a dog with a bone, but I don't have anything more to add! :)
 
Last edited:

solderdude

Grand Contributor
Joined
Jul 21, 2018
Messages
15,891
Likes
35,912
Location
The Neitherlands
This is why you'd compute an average for the responses of multiple headphones, with the Harman EQ curves applied.

If you believe that the Harman target and GRAS EQ curves derived from it are correct, then you should be able to reverse engineer a similar sounding target response curve for another type of measurement rig (like the 5128), by simply applying those Harman/GRAS-based EQ curves to the headphone data on the other rig. And then computing the average response of those EQ-corrected headphone measurements for the new rig.

The more headphone measurements you have for doing this, the more reliable this type of translation would probably be. The theory is totally sound though.

If you don't believe the Harman curve, and EQ curves derived from it are completely accurate though, then an approach like this would probably be mostly a waste of time and energy, and good plotting skills.

This is just my opinion, but I think sound power probably offers a much easier, and potential more accurate route to a neutral response curve for the 5128 than the above.

You can easily do this yourself.
Amir measured his HD650 on both the BK5128 and GRAS.
Good luck... If it were that easy it would have been done already and would have saved S. Olive the trouble of doing it himself.
 

ADU

Major Contributor
Joined
Jul 29, 2021
Messages
1,587
Likes
1,086
As I mentioned in my last post, Jaakopasenen already did this, but instead by comparing headphone measurements of many different models vs the same measurements of said models on another rig, this is how he came up with this "Harman Curves" for the various measurement rigs as a means of converting from GRAS. If you need to find out more, speak to Jaakopasenen about it. I don't have any more to add on your ideas of converting between measurement rigs, which I know you had enthusiasm for comparing Diffuse Field measurements.....I don't have anything to add on anything more in relation to these topics except - use GRAS measurements and the Harman Curve (Oratory / Amir / Crinacle / Resolve) and if your headphone has not been measured yet send it to Oratory to be measured, and then if all else fails use Jaakopasenen's site for his "Harman Curves" that are based on Innerfidelity measurements (ie not GRAS). I know you're like a dog with a bone, but I don't have anything more to add! :)

Understood.

I am fairly familiar with the approach Jaakko used to translate the Harman curve to different rigs' measurements. And believe you're correct that it was not done in precisely the same manner as what I've described above for the 5128 rig. The ideas are (somewhat) similar though.

Some of the other folk you mentioned above don't follow the Harman curve that religiously btw. I know Resolve doesn't in the upper treble, and believes as I do that the Harman curve is too rolled off and withdrawn in some spots there. And I believe he also uses a somewhat older version of the Harman curve for his target in the bass, that's a couple dBs lower than the current 2018 Harman over-ear target. I think Jaakko also did something similar on the bass level for his Harman-based EQ curves.

Using a lower target in the bass is probably not a bad idea for some open-back HPs. I'm less sure about this for closed-backs though.

I think Resolve might also possibly, maybe agree with me that the Harman curve could also be a bit too forward in the upper mids around 1.5 to 2k. But I'm less certain about his opinions on that. (So don't quote me on this.) :)

Haven't looked at any of Crin's EQ suggestions so far. Just some of his GRAS plots of over-ear HPs.
 
Last edited:

ADU

Major Contributor
Joined
Jul 29, 2021
Messages
1,587
Likes
1,086
In regards to Harman's research and the 5128, I believe that Sean Olive may have a word on the subject during the following conference :

Interesting. Thank you for sharing this, Maya.
 

Tadgh

Member
Joined
Aug 22, 2021
Messages
7
Likes
22
Location
Galway, Ireland
Genuine question, could one argue having an 89%/11% male/female split is a bit problematic? Would it not be better to have it closer to 50/50?
Just coming back to this because I think it's an interesting point that went under-addressed.

Given the average difference between male/female head and torso geometry and mass, as well as pinnae size - it's quite possible that the gender divide in Olive's testing did impact the Harman Curve's effectiveness against an "average" 50/50 population.

Of course, Olive rightfully pointed out that sound engineering is an intensely male industry and finding audio-science savvy participants to test different EQs on might have been more trouble than we imagine. Still, humans are varied and complex beings and I'd like to see the same tests run with larger sample sizes and broader demographics.
 

FireLion

Active Member
Joined
May 25, 2018
Messages
243
Likes
98
I don't have time for software EQ, I get a headphone or IEM and keep it if it sounds good, sell it if it's not. Always better for me when it can output the sound you want without tinkering.

All this arguing over headphone measurements! You guys all listen to music........right? ;)

1633324563496.png

Weird GB Flex!
 

Robbo99999

Master Contributor
Forum Donor
Joined
Jan 23, 2020
Messages
6,875
Likes
6,673
Location
UK
I don't have time for software EQ, I get a headphone or IEM and keep it if it sounds good, sell it if it's not. Always better for me when it can output the sound you want without tinkering.

All this arguing over headphone measurements! You guys all listen to music........right? ;)

View attachment 157085
Weird GB Flex!
Headphone EQ doesn't take any time if you use an existing preset for your headphone, for example from Oratory:
Although it does take some time initially to work out how to input the settings into whatever software you're using. Once you've set it up you don't ever have to think about it anymore, unless you enjoy experimenting with different EQ's or doing your own EQ experiments. EQ makes a big difference, so most people are gonna be missing out if they don't make the initial effort to get some EQ implemented.
 
Top Bottom