• 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!

10 Stages of Reading Headphone Measurements (by Resolve)

ADU

Major Contributor
Joined
Jul 29, 2021
Messages
1,584
Likes
1,086

This was a pretty fun list by Resolve. I think it may be a little more reflective of the phases he's personally been through in his own thinking on headphone measurements though, than some others. :) Though I can certainly identify with alot of what he's talkin about here. And I suspect some others here may as well. (Hence the reason for this post.)

There are some things I agree with in this, and others I don't though.

And I hope that stage 11 is realizing that we need more in-ear frequency response data on good-sounding loudspeakers in a room for comparison with the in-ear data of headphones. Preferably measured on the exact same anthropomorphic rigs (with a mannikin) that are used for the headphone FR measurements.
 
Last edited:

RickSanchez

Major Contributor
Cartographer
Joined
Sep 4, 2019
Messages
1,168
Likes
2,490
Location
Austin, TX
And I hope that stage 11 is realizing that we need more in-ear frequency response data on good-sounding loudspeakers in a room for comparison with the in-ear data of headphones. Preferably measured on the exact same anthropomorphic rigs (with a mannikin) that are used for the headphone FR measurements.

What's the benefit of this vs. using the the Klippel measurements? As Amir states (re: the Klippel measurement rig) in his speaker reviews:
Measurements that you are about to see were performed using the Klippel Near-field Scanner (NFS). This is a robotic measurement system that analyzes the speaker all around and is able (using advanced mathematics and dual scan) to subtract room reflections (so where I measure it doesn't matter). It also measures the speaker at close distance ("near-field") which sharply reduces the impact of room noise. Both of these factors enable testing in ordinary rooms yet results that can be more accurate than an anechoic chamber. In a nutshell, the measurements show the actual sound coming out of the speaker independent of the room.
 

NTK

Major Contributor
Forum Donor
Joined
Aug 11, 2019
Messages
2,656
Likes
5,819
Location
US East
I think @ADU is referring to this type of measurement. This was how Dr Olive's (Harman) headphone target curve got started. Picture is from Tyll Hertsens' trip to Harman with his HATS (head and torso simulator).

index.php
 

RickSanchez

Major Contributor
Cartographer
Joined
Sep 4, 2019
Messages
1,168
Likes
2,490
Location
Austin, TX
Got it, thanks for the context.

So please excuse my complete lack of knowledge in this area, as I ask the question a different way: what's the benefit of doing more of those types of measurements? How does this help us understand / improve on either speaker measurements or headphone measurements?

For example:
  • I've got the Klippel measurements for the Revel F228Be
  • I've got the Gras 45c measurements for the Sennheiser HD650 calibrated to 94 dBSPL @ 425 Hz
What would we gain by measuring the Revel F228Be again using an anthropomorphic rig like what's in the picture?
 
OP
A

ADU

Major Contributor
Joined
Jul 29, 2021
Messages
1,584
Likes
1,086
What's the benefit of this vs. using the the Klippel measurements? As Amir states (re: the Klippel measurement rig) in his speaker reviews:

The difference is that headphones are measured inside the ear. Whereas speakers usually are not.

So the microphone systems which are normally used for measuring the frequency response of a speaker will be missing all of the effects of the human body and the ear on the sound, before it actually reaches what is known as the ear drum reference point, or DRP (which is where most headphones are measured). This makes it much more difficult (but certainly not impossible) to do comparisons of, for example, the response of a good neutral loudspeaker in a room to a pair of headphones. To see if the latter is actually approximating something close to that speaker's in situ response.

One solution to this problem is to do measurements of the speakers in a room using the same kind of microphone system that is also used for doing headphone measurements. Which usually involves the use of a dummy head, with a mic installed inside the dummy's artificial ear at the DRP. This is the kind of setup which is shown in NTK's photo above.

Ideally this type of setup would also contain part of the dummy's torso as well, to capture some of the effects of a person's body on the speaker's sound field before it reaches the eardrum as well. But Tyll H. apparently didn't have one available for his system. So he used just the head and neck instead (which is all he used for making his headphone measurements).

There are some other methods of approximating a speaker's in-ear response than doing measurements from inside a dummy or human ear. (And Resolve actually mentions something about this in his stage 9 description in the video.)

One of the best and easiest methods of doing this, imho, is to simply combine the speaker's diffuse sound power response (from its spinorama graph) with the diffuse field response of the rig or dummy head being used for the headphone measurements. Convincing people that a method like this can actually work, and yield a decent result requires some actual in-ear measurement data from the speakers for comparison though. Which is something we don't really have for most of the rigs which are currently in use for measuring headphones. (Seeing is believing, as they say.)
 
Last edited:
OP
A

ADU

Major Contributor
Joined
Jul 29, 2021
Messages
1,584
Likes
1,086
I think @ADU is referring to this type of measurement. This was how Dr Olive's (Harman) headphone target curve got started. Picture is from Tyll Hertsens' trip to Harman with his HATS (head and torso simulator).

index.php

Thank you for digging this up, NTK. This is exactly the type of arrangement I had in mind (except that the dummy head is missing a torso).
 
Last edited:
OP
A

ADU

Major Contributor
Joined
Jul 29, 2021
Messages
1,584
Likes
1,086
Got it, thanks for the context.

So please excuse my complete lack of knowledge in this area, as I ask the question a different way: what's the benefit of doing more of those types of measurements? How does this help us understand / improve on either speaker measurements or headphone measurements?

For example:
  • I've got the Klippel measurements for the Revel F228Be
  • I've got the Gras 45c measurements for the Sennheiser HD650 calibrated to 94 dBSPL @ 425 Hz
What would we gain by measuring the Revel F228Be again using an anthropomorphic rig like what's in the picture?

As mentioned in my post above, it would help us to better compare the responses of the two transducers, and whether they appear to be more or less the same when actually measured inside of a human-like ear. Which might eventually help us to better understand and determine which headphones have a frequency response that is closest to a neutral loudspeaker.

This was Tyll Hertsen's objective as well. And it's why he accepted Harman's offer to let him do some in-ear measurements of loudspeakers with his own headphone measurement rig in Harman's reference listening room. He was trying to see what the frequency response of the speakers looked like when measured at the eardrum, to see how well they compared with the in-ear responses of the headphones he was measuring with the same dummy rig.

The room also plays an important role in this type of arrangement btw. Because most people create and listen to music in rooms, which are at least partially reflective. So ideally you would want that to also be reflected (figuratively speaking) in the speaker's in-ear measured response, just like you'd want the effects of the head, torso and ears included in the in-ear response. You could think of this as sort of the speaker's room+torso+head+ear-related transfer function (or R+HRTF), if you like. :)
 
Last edited:
OP
A

ADU

Major Contributor
Joined
Jul 29, 2021
Messages
1,584
Likes
1,086
Imo, the headphone frequency response data which is currently being captured by folks like Amir, Resolve, Oratory1990, Crin and other headphone reviewers and graphers would be more useful and probably somewhat easier to interpret if they also had some in-ear measurements of actual loudspeakers made on their rigs to use for comparison.

One problem with this though is that the rigs being used by many of these graphers for headphone measurements do not contain either a head (like Tyll's), or a torso. And they simply have the artificial pinna of the outer ear mounted on a flat plate. The GRAS 45CA rig used by ASR (and also Dr. Sean Olive) is one good example of this...

index.php


index.php


No head and torso means that you could not accurately capture the effects of the body and HRTF in a speaker's in-ear response when it's being measured in a typical semi-reflective room, like in the photo NTK posted.

This would be easier to do though with a measurement rig that includes an artificial head and torso, like the new HBK model, shown below, that Amir also tested...

index.php
 
Last edited:

Earfonia

Active Member
Joined
Feb 21, 2019
Messages
278
Likes
398
Location
Singapore

This was a pretty fun list by Resolve. I think it may be a little more reflective of the phases he's personally been through in his own thinking on headphones though, than some others. :) Though I can certainly identify with alot of what he's talkin about here. And I suspect some others here may as well. (Hence the reason for this post.)

There are some things I agree with in this, and others I don't though.

And I hope that stage 11 is realizing that we need more in-ear frequency response data on good-sounding loudspeakers in a room for comparison with the in-ear data of headphones. Preferably measured on the exact same anthropomorphic rigs (with a mannikin) that are used for the headphone FR measurements.

Stage 8 is exactly why I cannot use Harman Target for my IEM measurement. He is exactly right, HT is heavily smoothed and kinda useless for FR analysis above 5kHz, while we still have useful and important information in the 5-10kHz frequency response measurement detail. To smooth out those details is throwing away those useful info to the dust bin. I mentioned about it here in my measurement methodology:


And I prefer to set FR target in a more dynamic way. For example the ear canal resonance (primary resonance) at around 2.7kHz. The target should have some flexibility on that target frequency. IEM with measured primary resonance around 2.5kHz or 3kHz at around the same level would sound pretty much the same. If we just compensate that primary resonance at 2.7kHz base on target, they will look different.
 
OP
A

ADU

Major Contributor
Joined
Jul 29, 2021
Messages
1,584
Likes
1,086
@ADU Thanks for the detailed responses, I appreciate the info.

Not a problem. If you have other questions about anything above or in the video in the first post, feel free to ask, and I'll try to cobble together an opinion or two that will hopefully make a little sense. :) (Maybe Resolve will also drop by and give his take on some of this as well, which would be interesting).

One thing I should probably have mentioned is that the spinorama speaker data gathered by setups like Amir's Klippel system is still an important component in all this, even if you're doing the other in-ear measurements with a head and torso simulator rig. Because without the spinorama data or anechoic measurements, there's no real way of knowing if the speaker being used for these kinds of in-ear tests has a truly flat direct response, and a smooth linear-ish off-axis response. Because that is not the sort of info you can really get by looking at a speaker's in-ear response.

The torso, head, and ears on such headphone measurement rigs will all distort the frequencies of the speaker's in-ear response in such a way that a neutral response would no longer be represented by just a flat line, like it is for the direct response or listening window in a spinorama plot. So the in-ear data is more or less unusable as a tool for assessing the speaker's accuracy and neutrality in a normal listening space.

So both the in-ear, and the out-of-ear or anechoic measurements have their places in this. And are critical to understanding whether the speaker and headphones with a similar response are approximating something which could reasonably be called "neutral".
 
Last edited:

Earfonia

Active Member
Joined
Feb 21, 2019
Messages
278
Likes
398
Location
Singapore
I'm interested to see the differences of measurement result from various HATS (Head And Torso Simulator) rig. Sometime we assume that any HATS is a good representation of the average human HRTF. But I would like to see the measurement results from different HATS on the same measurement setup like shown in this picture:


Anyone has any info about that?

I know that HATS rig is ridiculously expensive so small chance to see measurement setup like the above done several time using different HATS from different manufacturers. Because HATS from different manufacturers might present different 'Neutral' target.
 

Earfonia

Active Member
Joined
Feb 21, 2019
Messages
278
Likes
398
Location
Singapore
And I found this:


Measuring HRTFs of Brüel & Kjær Type 4128-C, G.R.A.S. KEMAR Type 45BM, and Head Acoustics HMS II.3 Head and Torso Simulators:
 

Mad_Economist

Addicted to Fun and Learning
Audio Company
Joined
Nov 29, 2017
Messages
543
Likes
1,618
And I found this:


Measuring HRTFs of Brüel & Kjær Type 4128-C, G.R.A.S. KEMAR Type 45BM, and Head Acoustics HMS II.3 Head and Torso Simulators:
In addition to Snaidero et al, Hammershøi & Møller 2008 includes HRTF measurements of the KEMAR, HMSII.3, and 4128C, and some comparison and analysis vs. average population HRTFs.

Incidentally, acquiring more data on variation between anthropomorphic measurement fixtures is an interest point for me - I have a KEMAR head and a 4128, and someday I'm hoping to acquire a HMSII.3 (I was outbid at an auction semi-recently, to my sorrow), both for headphone and speaker-room system measurements.
 

Earfonia

Active Member
Joined
Feb 21, 2019
Messages
278
Likes
398
Location
Singapore
In addition to Snaidero et al, Hammershøi & Møller 2008 includes HRTF measurements of the KEMAR, HMSII.3, and 4128C, and some comparison and analysis vs. average population HRTFs.

Incidentally, acquiring more data on variation between anthropomorphic measurement fixtures is an interest point for me - I have a KEMAR head and a 4128, and someday I'm hoping to acquire a HMSII.3 (I was outbid at an auction semi-recently, to my sorrow), both for headphone and speaker-room system measurements.

Cool!

 

Mad_Economist

Addicted to Fun and Learning
Audio Company
Joined
Nov 29, 2017
Messages
543
Likes
1,618
Cool!

Sadly, the HATS on the right there - a vintage Neutrik-Cortex MK2, with blocked canal microphones - has been lost to me under duplicitous circumstances. I wish I'd done more measurements of it while I had it, but I can't really justify getting another one given that, in addition to the rarity, an IEC959 HATS just ain't that useful.
 
OP
A

ADU

Major Contributor
Joined
Jul 29, 2021
Messages
1,584
Likes
1,086
Stage 8 is exactly why I cannot use Harman Target for my IEM measurement. He is exactly right, HT is heavily smoothed and kinda useless for FR analysis above 5kHz, while we still have useful and important information in the 5-10kHz frequency response measurement detail. To smooth out those details is throwing away those useful info to the dust bin.

Thank you for the reply, Earfonia.

Imo, the most reliable way to get the sort of detail that Resolve and others want in the higher frequencies is to do some in-ear measurements of neutral speakers. The problem is that he and Headphonesdotcom do not have a suitable measurement rig for this purpose.

The next best approach, imo, is to compute an average response curve based on headphones which come the closest to a neutral response in the treble. This is essentially the approach I've been using with my DT-770's described here...


I would suggest also combining the sound power responses of some neutral loudspeakers with the diffuse field response of his rig. But there are some problems with that as well, since Resolve is just using a flat plate rig for his measurements. And consequently has no way of determining an accurate diffuse field HRTF for the rig.

For accurate diffuse field measurements, you would need a measurement rig with a head and torso. Just like you would need for in-ear measurements of speakers. And the approach with sound power + DF would still only be an approximation. (Though it would be a more accurate approximation than the Harman target imho.)

I mentioned about it here in my measurement methodology:


Very interesting! I'll have to take a little time to go through this info a little better though before offering comment on any of it.

And I prefer to set FR target in a more dynamic way. For example the ear canal resonance (primary resonance) at around 2.7kHz. The target should have some flexibility on that target frequency. IEM with measured primary resonance around 2.5kHz or 3kHz at around the same level would sound pretty much the same. If we just compensate that primary resonance at 2.7kHz base on target, they will look different.

All things being equal, an IEM should probably measure the same at the eardrum reference point as the in-ear response of a neutral speaker in a room to achieve roughly the same sound. This may not always work correctly though with the older 711 measurement couplers, because they are only designed to be accurate within a limited range of human hearing. And because they do not appear to emulate the length and other characteristics of an average human ear canal with great enough precision, as Jude explains here...

 
Last edited:
OP
A

ADU

Major Contributor
Joined
Jul 29, 2021
Messages
1,584
Likes
1,086
I'm interested to see the differences of measurement result from various HATS (Head And Torso Simulator) rig. Sometime we assume that any HATS is a good representation of the average human HRTF. But I would like to see the measurement results from different HATS on the same measurement setup like shown in this picture:


Anyone has any info about that?

I know that HATS rig is ridiculously expensive so small chance to see measurement setup like the above done several time using different HATS from different manufacturers. Because HATS from different manufacturers might present different 'Neutral' target.

Each manufacturer's HATS rig will respond similarly, but differently to an external stimulus. Because there are always some slight variations in their design and the other equipment that is used. So they will each have their own HRTF. (As the rigs become more accurate in their design though, these differences should gradually become less.)

If you were using one of the same rigs mentioned in your above link, or the ones in the PDF that Mad_Economist referenced, then you could take the diffuse field measurement of that specific rig, and combine it with the sound power responses of some neutral loudspeakers, and come up with something pretty close to a neutral in-ear response for the rig that way. It would only be an approximation though. (But still better than nothing at all, and probably also better than the Harman target imho.)

The same approach could also be used with the HBK 5128 HATS rig shown in one of my previous posts above, imo.
 
Last edited:
OP
A

ADU

Major Contributor
Joined
Jul 29, 2021
Messages
1,584
Likes
1,086
Dr. Olive also had some interesting thoughts on Resolve's 10 stages video, which can be found on his Twitter feed.
 
OP
A

ADU

Major Contributor
Joined
Jul 29, 2021
Messages
1,584
Likes
1,086
Since I brought up some of the shortcomings in the 711 system, I should also mention that in some recent studies, Harman researchers also uncovered some possible issues with the accuracy of the HBK's 5128 mannikin in terms of its potential leakage/interfacing effects with certain headphones. The study is discussed in the latter part of this video for a recent HBK conference. I've pre-cued the video to where that particular info begins. The rest is a summary of Harman's other work and findings related to headphones (so feel free to back it up to also give that a listen, if you wish.)


I believe there was some additional discussion on this subject in a couple topics here on ASR as well. This is one of them, but I think there may have been at least one other as well. (This is goin back several months now, so my memory is a bit fuzzy.)

 
Top Bottom