• WANTED: Happy members who like to discuss audio and other topics related to our interest. Desire to learn and share knowledge of science required as is 20 years of participation in forums (not all true). Come here to have fun, be ready to be teased and not take online life too seriously. We now measure and review equipment for free! Click here for details.

HBK Headphone Measurement Talks from Head-Fi and Sean Olive

ADU

Addicted to Fun and Learning
Joined
Jul 29, 2021
Messages
580
Likes
506
How is that different from the paper I presented at the HBK Seminar? I essentially took measurements of headphones that had been evaluated subjectively and measured on the GRAS45CA MOD (which you call JBL Modified) and measured the same headphones on a B&K 5128 to derive a target curve for that fixture. What would you do differently?

I can give you my opinion on that, if you want it (in private if you prefer, though I think others' feedback on this could also be quiet helpful). I don't think you'll like it though. :)

Hi, yes, I've still got that spreadsheet, attached! (@ADU )

Many thanks for this, Robbo!

I don't have a spreadsheet reader available at the moment. But will try to give this a look when I can.

Not to seem shrill... but there is really nobody who has an HE-1 Orpheus or is even interested in temporarily obtaining one in a professional capacity other than the paid or "sponsored" reviewers? The only real evidence that I can find of anyone owning one is a dealer claiming to be selling his personal unit in Australia, and a few random comments about how somebody knows some ultra-rich people in China who own a couple between them. Has nobody gone so far as to make an inquiry into a purchase? To obtain the very best in technology in any particular field is often not really possible, but here it seems doable, yet maybe that is just a mirage. It's just funny to me as someone from a visual background, you cannot feasibly get the best in display technology unless you are worth hundreds of millions or represent purchasing power of about as much, yet the best in audio reproduction can be had for the price of a used car.

A decent 65" LG OLED with close to the highest static contrast ratio in the home video market can be had for around 4 or 5 grand. Maybe you're referring to the higher end projectors though.

For the Harman work, "accuracy" is in reference to people's perception of the sound as being "accurate" or "uncoloured" as best as I understand it (@Sean Olive, your thoughts?), related to the long-running theme of "closing the circle of confusion". It's not accuracy in the sense of being accurate to an acoustic event - because, in part, there is no original acoustic event - but rather matching people's expectations of subjective timbre.

I think the latter is probably how quite a few audiophiles look at it though, Mad_Economist. Not that I'd really agree on that.
 
Last edited:

DualTriode

Addicted to Fun and Learning
Joined
Oct 24, 2019
Messages
520
Likes
292
Those are all good points.

<<<If we were starting from scratch we could use our box of test headphones, recorded sound test tracks and human preference data to develop a target curve for the current GRAS 45 CA-10, the HATS 5128 or some other test fixture. We could even use the older version JBL Modified GRAS 45 test fixture to develop a target curve. Each curve is specific to its own test fixture>>

How is that different from the paper I presented at the HBK Seminar? I essentially took measurements of headphones that had been evaluated subjectively and measured on the GRAS45CA MOD (which you call JBL Modified) and measured the same headphones on a B&K 5128 to derive a target curve for that fixture. What would you do differently?

Hello,

Thank you for the response.

I have not had the opportunity to see your paper that you presented at the HBK Seminar. I have only seen bits and pieces on this thread at ASR. You have no disagreement from me. Is a copy of your paper available?

“What would you do differently?”

I would do some additional experimentation.


I would derive a target curve for a flat panel test fixture with a surface mounted calibrated pressure microphone. It seems that there would be fewer extraneous variables to process in the ANOVA.


I am a retired ME with a APx555, GRAS 45CA-9 and an assortment of other calibrated GRAS microphones in my personal stash.

Thanks DT
 
Last edited:

DualTriode

Addicted to Fun and Learning
Joined
Oct 24, 2019
Messages
520
Likes
292
It is not free to experiment. Do you know the concept of a one-way function?

Hello,

My additional experimentation is to derive a target curve for a flat panel test fixture with a GRAS 46BD-S1 ¼” microphone.

The same procedure used by Dr. Olive to derive a target curve for his HATS 5128 test fixture.

Thanks DT
 

Sean Olive

Active Member
Audio Luminary
Technical Expert
Joined
Jul 31, 2019
Messages
217
Likes
1,769
Have you seen Chris Struck's paper on deriving in-room eardrum response from HRTFs, room acoustic parameters, and speaker directivity? I made a spreadsheet including the Harman room data at one point, I believe @Robbo99999 has it still.


Thank you Dr. Olive. I believe that Theile's position as of 2016 is that the DF target with preferential adjustments matching your adjustments to the in-room baseline of the Harman studies would be preferable for normal stereo content - that is, that the Harman target is correct :D


Thank you for the clarification - this is pretty much how I've been explaining it to folks when the Harman work comes up, do you mind if the bolded section is quoted as a single-line explanation of the Harman headphone target work?
I remember reading Chris's 2013 paper and glanced at it again yesterday. Very interesting. He took a different approach modelling the steady-state response of an anechoically flat loudspeaker with a given DI at the DRP accounting for the absorption characteristics of the room and contributions of direct vs reflected sound. The predicted target curves he comes up with seem to have more 3-4 kHz than the Harman Target and the bass is flat.

I have to re-read Theile's AES paper but I recall he didn't agree with the Harman Target. He argued that the bass shelf was an artifact of using pop music, and that it would be flat if we used orchestral (which we included).. He basic argument was that target curves are generalizations and eventually everything will be individualized to match our own ear canals and HRTFS. Maybe we read a different paper?
 

Mad_Economist

Senior Member
Technical Expert
Joined
Nov 29, 2017
Messages
425
Likes
1,194
I remember reading Chris's 2013 paper and glanced at it again yesterday. Very interesting. He took a different approach modelling the steady-state response of an anechoically flat loudspeaker with a given DI at the DRP accounting for the absorption characteristics of the room and contributions of direct vs reflected sound. The predicted target curves he comes up with seem to have more 3-4 kHz than the Harman Target and the bass is flat.
Indeed - Chris's paper, to my read, sets out to propose an alternative to the in-room measurement method for the baseline target response you adjusted based on listener preference
1634078873518.png

What strikes me as particularly intriguing about this is that, since anechoic data can be used to synthesize a DFHRTF, you could get a much higher resolution HRTF as a base than the (quite smoothed) in-room measurement.

I have to re-read Theile's AES paper but I recall he didn't agree with the Harman Target. He argued that the bass shelf was an artifact of using pop music, and that it would be flat if we used orchestral (which we included).. He basic argument was that target curves are generalizations and eventually everything will be individualized to match our own ear canals and HRTFS. Maybe we read a different paper?
I may have missed the portion where Theile argued that it would be flat with orchestral music. My read was that this portion of the conclusion to Theile 2016 was a concession that the Harman target is most preferred in general, with the caveat that Theile thinks that diffuse field is "flat and neutral" and thus that the Harman target is "less neutral" but more preferred:
1634079384268.png

My apologies if I'm paraphrasing it incorrectly, it had been a while since I read it in full as well :p
 

ADU

Addicted to Fun and Learning
Joined
Jul 29, 2021
Messages
580
Likes
506
What strikes me as particularly intriguing about this is that, since anechoic data can be used to synthesize a DFHRTF, you could get a much higher resolution HRTF as a base than the (quite smoothed) in-room measurement.

Not quite followin you on this, Mad_Economist. I think you'll have to expand a little more on what you mean by the above. I know what anechoic data, HRTFs, in-room measurements and so forth are though. So I'm followin you on that much. I'm not that much into binaural though, if that's what you're referring to.

From my perspective though, this is what equals a neutral response on a pair of headphones...

"a new target that simulates the in-room response of a high-quality loudspeaker system"

And not the curve shown in Figure 5 above. Which I believe is just a measurement rig's raw in-ear response to a pair of speakers which have been EQ-ed/calibrated to a flat frequency response at the listener's position.

Theile's remarks above are sort of reminiscent of the old-school thinking that a headphone should be tuned to match the response of a spectrally flat diffuse sound field, which would be much brighter-sounding. And probably more like listening to an anechoically flat speaker in an echo-free chamber or outdoors, without the low frequency boost you'd normally get in a more reflective space.

The Harman research largely disproved this old idea. And as Thiele seemed to mention in his comments, the headphone manufacturers never really followed the old diffuse field model anyway, even when it was considered "the standard".

"Neutral" in an anechoic chamber sounds very different than "neutral" in a typical semi-reflective room. And it is the latter of the two we want on a pair of headphones.
 
Last edited:

ADU

Addicted to Fun and Learning
Joined
Jul 29, 2021
Messages
580
Likes
506

For the above in-ear response curve (with the room gain removed) to be accurate, btw, it would have to have been done with a mannikin, to capture the head and torso-related effects, as well as the simulated ear's response to the flattened out sound of the two speakers in the room.

If they have a torso for the 5128, then Harman could easily repeat this type of test for their new rig. And then combine the result with the estimated in-room responses of some anechoically flat speakers to derive their new neutral reference curve for the 5128 rig.

I don't really see why you'd bother going to all that additional trouble though, when you could simply measure the in-ear response of two anechoically flat speakers in a semi-reflective room "as is", without EQ-ing them to remove the room gain. And then use that as your target. If you want to see what the in-ear response of a pair of neutral, anechoically flat speakers looks like when they are in a typical semi-reflective listening space, then why not just measure that, rather than making the process more difficult by adding the other steps above?

Maybe I'm missing something though. And it's easier somehow to derive the target the other way. (?)
 
Last edited:

Mad_Economist

Senior Member
Technical Expert
Joined
Nov 29, 2017
Messages
425
Likes
1,194
Not quite followin you on this, Mad_Economist. I think you'll have to expand a little more on what you mean by the above. I know what anechoic data, HRTFs, in-room measurements and so forth are though. So I'm followin you on that much. I'm not that much into binaural though, if that's what you're referring to.

Right, so the baseline of the Harman target is an in-room measurement of a Revel F208, in the Harman room, which had been equalized to a flat response at the listening position on an omni mic, measured with a GRAS 45CA, right? The original (well, "original" - 2013) Harman target was derived by applying a pair of shelf filters based on listener preferences to that baseline
1634088055596.png

This, clearly, generally works rather well, since it's the foundation of the Harman work, which predicts subjective preference well. However, because it's an in-room measurement, to achieve a smooth target, a fairly large amount of smoothing was required (@Sean Olive can correct me if I'm wrong, but people less directly connected have said it's in the 1/2-1/4 octave range). This obscures some finer HRTF features and slightly changes others (such as the shape of the rise into the "ear gain" region).

These small details don't matter as much as the general trend, which of course is why the Harman target is overwhelmingly preferred, bur being able to work with higher resolution data is handy because headphone response measurements typically show features that are much high Q than the in-room measurement's resolution can incorporate. With GRAS fixtures, the "pinna notch" feature around 9khz is the iconic example - B&K gear typically has a constructive (rather than destructive) feature in this same area. It would improve our targets to have these small details included, and with anechoic data, we can get very high resolution HRTFs to work without having room modes or reflections interfering.

From my perspective though, this is what equals a neutral response on a pair of headphones...

"a new target that simulates the in-room response of a high-quality loudspeakers system"

And not the curve shown above. Which I believe is just a measurement rig's raw in-ear response to a pair of speakers which have been EQ-ed/calibrated to a flat frequency response at the listener's position.
What "neutral" means is frankly something I find kind of uninteresting, simply because it descends into pedantry where everyone has a different internal definition they're nitpicking towards. Theile's definition is based on his hearing model and binaural sound propagation. The version that I usually understand Sean to be using is in reference to the perception that the sound is uncoloured by subjective listeners. Neither of these definitions seems unreasonable to me, and I don't see much use in arguing about which one is more right, but rather examining what, other than, "being neutral", makes these compelling pitches.

Theile's remarks above are sort of reminiscent of the old-school thinking that a headphone should be tuned to match the response of a spectrally flat diffuse sound field, which would be much brighter-sounding. And probably more like listening to an anechoically flat speaker in an echo-free chamber or outdoors, without the low frequency boost you'd generally get in a more reflective space.
That's not overly surprising, given that Theile's work underpins the DF target. However there is a quite substantial difference in sound between FF (a frontally located flat loudspeaker in an anechoic environment) and DF - and indeed attacking FF both on its inaccuracy under his paradigm of reproduction and its subjective non-preferability is what Theile 1986 is about.

Regarding "low frequency boost", I feel like I get into this dialogue about every 5 days at this point, but an anechoically flat loudspeaker's in-room power response doesn't show a Harman-esque bass shelf unless it maintains directivity control to an atypically low frequency, and then rapidly transitions to omnidirectionality. The Revels used track the general 1dB/oct or so slant pretty well without equalization in-room
1634088962579.png

but their slant is closer to the "B&K curve". As, of course, there's a degree of circularity in defining speakers as good based on anechoically flat response and power response that yields the preferred in-room power response (would we more prefer a non-anechoically-flat speaker with differing directivity but the same power response?).

For the above in-ear response curve (with the room gain removed) to be accurate, btw, it would have to have been done with a mannikin, to capture the head and torso-related effects, as well as the simulated ear's response to the flattened out sound of the two speakers in the room.
The impacts of shoulders are less than you might think in an environment with reflections, but yes, I believe the 43AG used in an earlier paper was placed in a mannequin for specifically this reason.

I don't really see why you'd bother going to all that additional trouble though, when you could simply measure the response of the two anechoically speakers in the semi-reflective room "as is", without equalizing them to remove the room gain... if that's what you really want? If you want to see what the in-ear response of a pair of neutral anechoically flat speakers looks like when they are in a typical semi-reflective listeneing space, then why not just measure that, rather than making the process more compliacted by adding the other steps above?
I'm unsure what "the other steps above" are, in this case - if you're referring to the Harman work in the period where in room measurements were being done (e.g. Listener Preferences for Different Headphone Target Response Curves and Listener Preferences for In-Room Loudspeaker and Headphone Target Responses), IMO the initial flattening of the speaker in situ response serves an interesting function in showing that the preferred EQ adjustments for both headphones and speakers are similar - that is, that preferred headphone frequency response and speaker power response are roughly equivalent.
 
Last edited:

thewas

Major Contributor
Forum Donor
Joined
Jan 15, 2020
Messages
3,390
Likes
7,137
Regarding "low frequency boost", I feel like I get into this dialogue about every 5 days at this point, but an anechoically flat loudspeaker's in-room power response doesn't show a Harman-esque bass shelf unless it maintains directivity control to an atypically low frequency, and then rapidly transitions to omnidirectionality. The Revels used track the general 1dB/oct or so slant pretty well without equalization in-room
1634088962579.png

but their slant is closer to the "B&K curve". As, of course, there's a degree of circularity in defining speakers as good based on anechoically flat response and power response that yields the preferred in-room power response (would we more prefer a non-anechoically-flat speaker with differing directivity but the same power response?).
We shouldn't also forget that the F208 is only anechoically relatively flat till approximately 50 Hz and then its response drops:

Spin%2B-%2BRevel%2BPerforma3%2BF208.png

Source: https://speakerdata2034.blogspot.com/2019/03/spinorama-data-revel-home.html?view=classic

Agree though that the response of anechoically flat loudspeaker in the bass is closer to the B&K target, Dr. Toole also questioned and tried some possible explanations why the preferred levels at both frequency extremes differed to the anechoic ones and said further research must be done:

The “all listeners” average curve
is close to the predicted target, except at low frequencies
where it is apparent that the strongly expressed preferences
of inexperienced listeners significantly elevated the aver-
age curve. In fact, the target variations at both ends of the
spectrum are substantial, with untrained listeners simply
choosing “more of everything.” An unanswered question
is whether this was related to overall loudness—more re-
search is needed. However, most of us have seen evidence
of such more-bass, more-treble listener preferences in the
“as found” tone control settings in numerous rental and
loaner cars.
More data would be enlightening, but this amount is
sufficient to indicate that a single target curve is not likely
to satisfy all listeners. Add to this the program variations
created by the “circle of confusion” and there is a strong
argument for incorporating easily accessible bass and treble
tone controls in playback equipment. The first task for such
controls would be to allow users to optimize the spectral
balance of their loudspeakers in their rooms, and, on an
ongoing basis, to compensate for spectral imbalances as
they appear in movies and music.
The attenuated high frequencies preferred by the trained
listeners stands in contrast to the preferences exhib-
ited by those same listeners in numerous double-blind
multiple-comparison loudspeaker evaluations. In those
tests, it is the flat on-axis loudspeakers that are most highly
rated (those that perform close to the predicted curve in
Fig. 14). Is this a consequence of the different experimental
methods: the different listener tasks? In one, listeners ad-
justed the bass and/or treble balance in a single loudspeaker
model; in the other they rated spectral balances and other at-
tributes in randomized comparisons of different products. It
is a subtle but important difference awaiting an explanation.


Source AES paper 17839 "The Measurement and Calibration of Sound Reproducing Systems": https://www.aes.org/e-lib/browse.cfm?elib=17839 (Open access free download)
 

Robbo99999

Major Contributor
Forum Donor
Joined
Jan 23, 2020
Messages
3,188
Likes
2,703
Location
UK
Right, so the baseline of the Harman target is an in-room measurement of a Revel F208, in the Harman room, which had been equalized to a flat response at the listening position on an omni mic, measured with a GRAS 45CA, right? The original (well, "original" - 2013) Harman target was derived by applying a pair of shelf filters based on listener preferences to that baseline
View attachment 158700
This, clearly, generally works rather well, since it's the foundation of the Harman work, which predicts subjective preference well. However, because it's an in-room measurement, to achieve a smooth target, a fairly large amount of smoothing was required (@Sean Olive can correct me if I'm wrong, but people less directly connected have said it's in the 1/2-1/4 octave range). This obscures some finer HRTF features and slightly changes others (such as the shape of the rise into the "ear gain" region).

These small details don't matter as much as the general trend, which of course is why the Harman target is overwhelmingly preferred, bur being able to work with higher resolution data is handy because headphone response measurements typically show features that are much high Q than the in-room measurement's resolution can incorporate. With GRAS fixtures, the "pinna notch" feature around 9khz is the iconic example - B&K gear typically has a constructive (rather than destructive) feature in this same area. It would improve our targets to have these small details included, and with anechoic data, we can get very high resolution HRTFs to work without having room modes or reflections interfering.


What "neutral" means is frankly something I find kind of uninteresting, simply because it descends into pedantry where everyone has a different internal definition they're nitpicking towards. Theile's definition is based on his hearing model and binaural sound propagation. The version that I usually understand Sean to be using is in reference to the perception that the sound is uncoloured by subjective listeners. Neither of these definitions seems unreasonable to me, and I don't see much use in arguing about which one is more right, but rather examining what, other than, "being neutral", makes these compelling pitches.


That's not overly surprising, given that Theile's work underpins the DF target. However there is a quite substantial difference in sound between FF (a frontally located flat loudspeaker in an anechoic environment) and DF - and indeed attacking FF both on its inaccuracy under his paradigm of reproduction and its subjective non-preferability is what Theile 1986 is about.

Regarding "low frequency boost", I feel like I get into this dialogue about every 5 days at this point, but an anechoically flat loudspeaker's in-room power response doesn't show a Harman-esque bass shelf unless it maintains directivity control to an atypically low frequency, and then rapidly transitions to omnidirectionality. The Revels used track the general 1dB/oct or so slant pretty well without equalization in-room
View attachment 158707
but their slant is closer to the "B&K curve". As, of course, there's a degree of circularity in defining speakers as good based on anechoically flat response and power response that yields the preferred in-room power response (would we more prefer a non-anechoically-flat speaker with differing directivity but the same power response?).


The impacts of shoulders are less than you might think in an environment with reflections, but yes, I believe the 43AG used in an earlier paper was placed in a mannequin for specifically this reason.


I'm unsure what "the other steps above" are, in this case - if you're referring to the Harman work in the period where in room measurements were being done (e.g. Listener Preferences for Different Headphone Target Response Curves and Listener Preferences for In-Room Loudspeaker and Headphone Target Responses), IMO the initial flattening of the speaker in situ response serves an interesting function in showing that the preferred EQ adjustments for both headphones and speakers are similar - that is, that preferred headphone frequency response and speaker power response are roughly equivalent.
This little EQ experiment of mine might be interesting in relation to the interrelationship of the baseline in room flat response in headphones vs their speaker Harman Curve vs their headphone Harman Curves.

Step #1: I EQ'd my K702 headphones to the baseline in room flat response (taken from the Harman Data, your spreadsheet you gave me):
K702 to Harman In Room Flat.jpg


Step #2: I applied a 9dB Linear Tilt (which is the last 3 filters) to the result of the above from 10Hz-20kHz, which is now essentially simulating an anechoic speaker in a regular room:
Harman Flat with 9dB Tilt.jpg


Step #3: I compared the result from Step#2 with the 2013 Harman Curve and it just so happens it's a perfect match in the treble with just the bass being slightly underrepresented:
Harman Flat with 9dB Tilt vs Harman 2013.jpg


I'm a bit unsure if I like it better than the 2018 Harman Curve which is my long time favourite, but the Linear Tilt sounds the best at 9dB tilt, so in terms of me just adjusting the Linear Tilt then this is optimised. However, I tried the above with a further Low Bass Shelf boost (below 100Hz) and it does help, but then it can cloud the rest of the frequency response which I think is what Harman were thinking by increasing the treble in the 2018 Harman Curve to offset that effect, along with cutting out a bit of the response in the 200Hz region. I would say my result in Step #3 above is good for Virtual 7.1 Surround gaming, it does feel praps a bit more natural & accurate in that environment, and it's ok for music, but I feel it does lack a bit of bass & treble for music listening - I think perhaps the best headphone target curve is not necessarily the most "accurate" one, probably because headphone listening doesn't have that tactile bass you get from speakers (and once you've increased the bass then you have to increase the treble to bring back the clarity ie 2018 Harman Curve - my intuition).
 
Last edited:

DualTriode

Addicted to Fun and Learning
Joined
Oct 24, 2019
Messages
520
Likes
292
Hello @amirm, Mad_Economist and All,

This thread seems to have everyone referencing and quoting AES Convention Papers, I will join in as well.

You folks appear to have the opinion that headphone measurement and target curves cannot be done without a simulated ear test fixture.

In an effort to find out if anyone had researched and ruled out flat plate headphone measurement because of lack of using Drum Reference Point or some related issue I searched the Library at AES.org. It turns out searching aes.org finds some interesting results, including:

U. Horbach, "Characterizing the Frequency Response of Headphones—A New Paradigm," Paper 9274, (2015 May.). doi:

By Harman’s own Ulrich Horbach

Here are a couple of fair use quotes:

>>> “ this leads to the conclusion that humans largely eliminate the influence of their external ear features by inverse filtering in the brain. We should therefore attempt to characterize acoustic sources such as headphones before the sound enters the pinna.” From page #1 introduction

“Despite common assumptions, it seems of less importance to terminate the headphone with a correct acoustic load. Plate and array methods yield very similar curves (Fig. 9 lower graph). The ear simulator test is the result of a mix between the headphone under test and the frequency response of the coupler itself, and contains details apparently unrelated to the perceived response (Fig. 8).” From page #4 under heading 4. RESULTS AND DISCUSSION

Looks like I am not the first to think that measuring headphones with a plate measurement fixture may return superior results.



Thanks DT
 

ADU

Addicted to Fun and Learning
Joined
Jul 29, 2021
Messages
580
Likes
506
This little EQ experiment of mine might be interesting in relation to the interrelationship of the baseline in room flat response in headphones vs their speaker Harman Curve vs their headphone Harman Curves.

Step #1: I EQ'd my K702 headphones to the baseline in room flat response (taken from the Harman Data, your spreadsheet you gave me):


Step #2: I applied a 9dB Linear Tilt (which is the last 3 filters) to the result of the above from 10Hz-20kHz, which is now essentially simulating an anechoic speaker in a regular room:


Step #3: I compared the result from Step#2 with the 2013 Harman Curve and it just so happens it's a perfect match in the treble with just the bass being slightly underrepresented:

I'm a bit unsure if I like it better than the 2018 Harman Curve which is my long time favourite, but the Linear Tilt sounds the best at 9dB tilt, so in terms of me just adjusting the Linear Tilt then this is optimised. However, I tried the above with a further Low Bass Shelf boost (below 100Hz) and it does help, but then it can cloud the rest of the frequency response which I think is what Harman were thinking by increasing the treble in the 2018 Harman Curve to offset that effect, along with cutting out a bit of the response in the 200Hz region. I would say my result in Step #3 above is good for Virtual 7.1 Surround gaming, it does feel praps a bit more natural & accurate in that environment, and it's ok for music, but I feel it does lack a bit of bass & treble for music listening - I think perhaps the best headphone target curve is not necessarily the most "accurate" one, probably because headphone listening doesn't have that tactile bass you get from speakers (and once you've increased the bass then you have to increase the treble to bring back the clarity ie 2018 Harman Curve - my intuition).

Apologies if this is stepping a bit on DualTriode's points, which I think are also interesting... but a -0.9 dB per octave tilt sounds about right, Robbo.

I haven't played around as much with the estimated in-room response curves in the spinorama data as with the sound power data. But I think the average for an in-room response is generally somewhere around the -0.75 to -1.0 (or -1.25?) dB per octave range, for a neutral loudspeaker with a flat direct/on-axis response. Since there are ten octaves in the normal human hearing range of 20 to 20000 Hz, simply multiple those dB/octave values by 10 to get the total offset in dBs at 20000 Hz. (Which would be the -9 dB tilt in your above example.) Don't quote me on those figures though.

The sound power response curves have a somewhat steeper tilt though which is generally more in the -1.0 to -1.5 dB per octave range.

If you compare either the in-room or sound power responses of speakers to a constant slope in these ranges, I think you will also find that there is indeed a little bit of boost in both the bass and treble relative to the midrange, particularly the upper mids, where there's often even a little more of a drop at the cross-over of the midrange/tweeter junction at ~2 kHz. Speakers with less well-designed cross-overs will often have reduced dispersion (aka greater directivity) in that range, which results in a bit more of dip in their in-room responses there.

Some sound power responses of various groups of speakers (explained below) versus a constant -1.26 dB per octave slope.

index.php


index.php


These variations in amplitude from a constant slope will probably be a little more pronounced though in the estimated sound power curves than in the estimated in-room curves. Because the latter incorporates more of a speaker's direct/on-axis response.

This is the average sound power response of the ten V-shaped speakers shown in the first graph above, with their approximately -1.26 dB per octave overall slope restored...

10VSHAPEDSPSCALED.jpg


The speakers in this next average sound power graph have what I would describe as more of an "L-shaped" response. They differ a bit from the more V-shaped speakers above, because they don't rise back up as much in the higher frequencies above 2k. And tend to form more of a plateau (rather than a foothill) above that frequency range...

10LSHAPEDSPSCALED.jpg


This is also not an uncommon type of response for some speakers, including JBLs, some Neumanns (KH 80, KH 120A, KH 310A), and also a few Genelecs (1032A and 8050B).

Fwiw, there are also a fair number of speakers that have a noticeably more linear SP response than any of the examples shown above. But those tend to be less common (and in some cases, also quite a bit more expensive) than some of the types shown above.

All of these different types of speakers could probably be considered "neutral" since they can all be quite flat in their direct/on-axis response. But they would all sound slightly different when put into a room, because of their different directivity and off-axis responses.
 
Last edited:

Robbo99999

Major Contributor
Forum Donor
Joined
Jan 23, 2020
Messages
3,188
Likes
2,703
Location
UK
Apologies if this is stepping a bit on DualTriode's points, which I think are also interesting... but a -0.9 dB per octave tilt sounds about right, Robbo.

I haven't played around as much with the in-room response curves in the spinorama data as with the sound power data. But I think the average for an in-room response is generally somewhere around the -0.75 to -1.0 (or -1.25?) dB per octave range, for a neutral loudspeaker with a flat direct/on-axis response. Since there are ten octaves in the normal human hearing range of 20 to 20000 Hz, simply multiple those dB/octave values by 10 to get the total offset in dBs at 20000 Hz. (Which would be the -9 dB tilt in your above example.) Don't quote me of those figures though.
About the 9dB Tilt, I also dug up a in-room measurement from last year of my JBL 308p after Anechoic EQ - following is an average of a number of measurements at 2m listening position, it also follows the Harman Curve quite naturally which is the same 9dB from peak of the bass to treble at 20kHz (ignore the fact I applied a -5dB Low Shelf, it was when I used to run my speakers with a +5dB Low Shelf & subsequent RoomEQ so I had to subtract that Low Shelf to show the untouched Anechoic EQ):
Anechoic EQ.jpg

and here's a more recent measurement in room but just one measurement (not average) of my latest Anechoic EQ which I finetuned, I'd done this measurement for distortion purposes but it shows the slope of the frequency response:
JBL 308p distortion 74dB 2m 2speakers (dB).jpg

If you compare either the in-room or sound power responses of speakers to a constant slope in these ranges, I think you will also find that there is indeed a little bit of boost in both the bass and treble relative to the midrange, particularly the upper mids, where there's often even a little more of a drop at the cross-over of the midrange/tweeter junction at ~2 kHz. Speakers with less well-designed cross-overs will often have reduced dispersion (aka greater directivity) in that range, which results in a bit more of dip in their in-room responses there.


These variations in amplitude from a constant slope will probably be a little more pronounced though in the estimated sound power curves than in the estimated in-room curves. Because the latter incorporates more of a speaker's direct/on-axis response.

So you're saying that the slight V-shaped nature of a lot of speakers might be related to the preference for the Harman 2018 curve which has slightly more bass and slightly more treble than an otherwise neutral in room response - as in a lot of music may have been produced on slightly V-shaped speakers so therefore this fits the slightly V-shaped nature of the 2018 Harman Headphone Curve? Perhaps that's part of it, I was thinking more along the lines that headphones needed a bit more bass due to the loss of the tactile bass which you get from speakers, and then if you're increasing the bass beyond the natural level then I was thinking you need to balance that with a bit more treble to balance it - so that was my thinking about the preference for the slightly V-shaped Harman 2018 Curve.
 
Last edited:

Sean Olive

Active Member
Audio Luminary
Technical Expert
Joined
Jul 31, 2019
Messages
217
Likes
1,769
Thank you for your post above, Dr. Olive.

I'd like to see more good spinorama type measurements of well-extended speakers with either separate or built-in sub-woofers. Because there aren't really enough at the moment to draw precise conclusion on all of this. As I've previously tried to point out though (both here, and now also here), based on the spinorama data that is currently available, I see no appreciable differences between your preferred in-room loudspeaker curve or PIRL (and the current Harman over-ear headphone targets based to some extent on this), and the estimated in-room responses of some of the best extended speakers in the sub-bass, which also have a reasonably flat/neutral on-axis response...

Dutch & Dutch 8C: https://pierreaubert.github.io/spinorama/Dutch Dutch 8C/ErinsAudioCorner/index_eac.html
KEF Reference 5: https://pierreaubert.github.io/spinorama/KEF Reference 5/KEF/index_vendor.html
Infinity Prelude MTS: https://pierreaubert.github.io/spinorama/Infinity Prelude MTS/Infinity/index_vendor.html
Mesanovic RTM10: https://pierreaubert.github.io/spinorama/Mesanovic RTM10/Mesanovic/index_vendor.html
Infinity Intermezzo 4.1t: https://pierreaubert.github.io/spinorama/Infinity Intermezzo 4.1t/Infinity/index_vendor.html

All that suggests to me is that people prefer the sound of really good speakers, that are either well-extended into the sub-bass due to an on-board sub-woofer and/or calibration. Or accompanied by a good standalone sub (or subs). Given the recent proliferation of home theaters with these kinds of setups, I suppose this sort of preference isn't really that surprising. If you feel the need to attribute this to something else (like an absence of tactility), then so be it I guess. But I really see no necessity for that, based on looking at the estimated in-room measurements of these kinds of speakers. The measurements of the two seem to be tracking pretty well, without any other explanations.

All of the speakers in the above links are better extended in the sub-bass than the Revel F208 used for some of your headphone tests...

Revel F208 (ASR measurement V2): https://pierreaubert.github.io/spinorama/Revel F208/ASR/index_asr-v2-20210519.html
Revel F208 (ASR measurement V1): https://pierreaubert.github.io/spinorama/Revel F208/ASR/index_asr-v1-20200508.html
Revel F208 (Revel measurement) https://pierreaubert.github.io/spinorama/Revel F208/Revel/index_vendor.html

And they are also pretty flat and linear in their direct/on-axis response. So they would all fall into the "neutral" category imo.

They are probably pretty close to the top of the heap in terms of their extension into the lower frequencies though. And just to be totally fair, most of the measurements shown above were created by the vendors (including Harman in the case of the Infinitys), rather than other independent sources (such as ASR or Erin's Audio Corner). And the Infinity Intermezzo 4.1t does "cheat" the direct response up a bit in the sub-bass. And the KEF (which has an unusually linear off-axis response) also appears to be tilted a bit more in favor of the lower frequencies.

The better extended a speaker is in the sub-bass though, the better it seems to track with the more elevated levels in the sub-bass in your target response curves for both loudspeakers and headphones. So as already mentioned, I don't see much need beyond this to better explain the levels of the preferred response curves in that range.
<<<I see no appreciable differences between your preferred in-room loudspeaker curve or PIRL (and the current Harman over-ear headphone targets based to some extent on this), and the estimated in-room responses of some of the best extended speakers in the sub-bass, which also have a reasonably flat/neutral on-axis response...>>

I agree. And that was the point of showing the anechoic spins of the Revel vs the in-room and headphone target curves that you show in the next post. The bass roll-off of the speaker only means that a subwoofer or EQ needs to be applied.
 

Sean Olive

Active Member
Audio Luminary
Technical Expert
Joined
Jul 31, 2019
Messages
217
Likes
1,769
Indeed - Chris's paper, to my read, sets out to propose an alternative to the in-room measurement method for the baseline target response you adjusted based on listener preference
View attachment 158681
What strikes me as particularly intriguing about this is that, since anechoic data can be used to synthesize a DFHRTF, you could get a much higher resolution HRTF as a base than the (quite smoothed) in-room measurement.


I may have missed the portion where Theile argued that it would be flat with orchestral music. My read was that this portion of the conclusion to Theile 2016 was a concession that the Harman target is most preferred in general, with the caveat that Theile thinks that diffuse field is "flat and neutral" and thus that the Harman target is "less neutral" but more preferred:
View attachment 158682
My apologies if I'm paraphrasing it incorrectly, it had been a while since I read it in full as well :p
Thanks for reposting this paper by G. Theile.

I forgot I wrote this quote:

“While most loudspeaker manufacturers today aim to achieve a flat frequency response on-axis, headphone manufacturers seem to be aiming at a target response that is as variable and random as the weather.” ( I was thinking Canadian weather --not southern California weather.) I think CA Poldy said much the same thing many years earlier.

If we repeated our study where different targets including the DF ones were rated by trained listeners according to neutrality or naturalness rather than preference. I don't think the results would be much different.

His question about "whether the majority of companies would agree" to adopting the Harman Target is intriguing. I think some already have, but others will continue to go their own way until there is a better alternative or the industry moves closer towards personalization.

It is clear based on our surveys of headphone measurements and Breebarts' (2017) that the majority of companies are not adopting the IEC DF standard. The questions is why? There are at least three independent studies I am aware of (ours, Lorho and Fleischmann) that evaluated headphones equalized to DF vs other targets and found the latter to be preferred.
 
Last edited:

Robbo99999

Major Contributor
Forum Donor
Joined
Jan 23, 2020
Messages
3,188
Likes
2,703
Location
UK
@ADU @Mad_Economist , rather than using a purely linear tilt to apply to Harman Headphone Flat In-Room Baseline measurement I've decided to try applying their Harman Modified In-Room Loudspeaker Curve pictured as the red dotted line below:
index.php

So the red dotted line is not perfectly linear and does provide a bit more of a boost below 100Hz which is where I felt it was needed from my headphone listening experiment in my last post. I decided to accurately trace the red dotted line in VirtuixCAD and then work out a set of parametric filters that describes this tilt from a flat response, following are a few comparisons.

The Linear Tilt Filters I was using don't fully describe the Modified In-Room Loudspeaker Target, more bass boost in the target:
Linear Tilt vs Idealised Harman Target.jpg


Following are the set of parametric filters that are required to convert a flat in-room response to the Harman Modified In-Room Loudspeaker Target:
Harman Modified In-room Speaker Curve Filters.jpg


Find attached at end of post the EqualiserAPO config file that included these filters. I thought you might find it useful in your musings/experimentations. I'm gonna apply this Harman Modified In-Room Loudspeaker Target to my K702 that has been EQ'd to the baseline flat in-room response, following pic is the baseline flat in-room response:
index.php

And following is with the Harman Modified In-Room Loudspeaker Target applied to the above:
Harman Modified In-room Speaker Curve applied to Harman Flat In-Room Baseline Headphone vs Har...jpg

Now that last pic there is crazy, notice how it just happens to follow the 2013 Harman Headphone Curve to absolute perfection apart from having a bit too much energy in the 100-200Hz zone!! (I thought Harman created this curve by listeners tweaking treble & bass knobs around the baseline target, but it totally tracks the theory of applying the Harman Modified In-Room Loudspeaker Curve to the Harman Baseline In-Room Flat Headphone measureement.) It's probably safe to say that the curve above in the last pic is the most neutral/accurate curve you could expect in a headphone as a generalised curve! It's essentially the Harman 2013 curve, I haven't listened to the above slight variation on the Harman 2013 Curve, but I will do.

EDIT: listened with the curve pictured in the last pic which is essentially the 2013 Harman Curve, and it sounds great, this might become my target curve of choice, I've chosen to keep the extra energy between 100-200Hz - and remember this is actually the Harman Baseline Flat In-Room Headphone Measurement combined with Harman Modified In-Room Loudspeaker Target so theoretically is the most neutral/accurate generalised target curve out there.
 

Attachments

  • Harman Modified In-Room Loudspeaker Target.txt
    759 bytes · Views: 2
Last edited:
OP
amirm

amirm

Founder/Admin
Staff Member
CFO (Chief Fun Officer)
Joined
Feb 13, 2016
Messages
35,409
Likes
136,154
Location
Seattle Area
With GRAS fixtures, the "pinna notch" feature around 9khz is the iconic example - B&K gear typically has a constructive (rather than destructive) feature in this same area. It would improve our targets to have these small details included, and with anechoic data, we can get very high resolution HRTFs to work without having room modes or reflections interfering.
Why? Your ERB is already getting quite large by those frequencies so those ups and downs are not heard. You can easily experiment with this by creating high Q filters and listening for effects. I have done this countless times and they simply are not audible.

That is on top of the problem of small variations of headphone on the fixture can vary these reflections and resulting vector responses.
 
Top Bottom