[ADUHF]

Sure, @ADUHF, I'll offer some detailed thoughts when I'm back in the office. (I'm on a much-needed vacation.)

Hope you're enjoying your time off. (And not thinking too much about more trivial stuff like headphone measurements and response curves, if that's possible.) That would be great though, whenever you get the chance. Since I didn't hear Sean's presentation and have never used either the 5128 or GRAS rigs myself, it would be helpful to hear some other opinions on the differences from folks like yourself who have.

It would not surprise me if there are a few wrinkles that might need to be ironed out in the 5128 system, since this is still a very new system compared to the 711 systems (like the GRAS) which have been in use for literally decades. And perhaps Sean and the folks at Harman have identified one of those wrinkles.

I know that HBK put quite a bit of effort and research into designing their new system though. And would expect there to be some differences in the way it measures from the GRAS, since the designs of the two systems are so different. So I guess that's why I'm not overly surprised by some of the results.

I think it'd also be helpful to see FR graphs of other headphones that Harman compared on the two rigs though, to get a more general idea of the differences. Though I suspect there are similar plots of some other headphones that have been measured on both, which could maybe be posted here for comparison as well.

No, of course I don't mind you sharing this here.

Cool.

 

[ADUHF]

This is probably neither here nor there. But just for grins and giggles, I did a few comparison plots of the diffuse field compensation curves of the two rigs, to see what the differences look like on those.

Based on looking at various measurements on the two rigs though, I have alot more faith in the accuracy of the 5128's DF compensation curve than in the DF curves I've run across so far for the GRAS.... So I don't think these plots really mean a whole lot. I'll go ahead and post them anyway though. (And hopefully they won't just add to some people's confusion.) The vertical divisions on these graphs are all in steps of 1 dB.

The first comparison is between the 5128 DF curve, and the DF curve which is currently used in Oratory's Graphing Tool...

Next is a comparison of the 5128 DF to the DF curve that Oratory was using in his Graphing Tool prior to this year (so pre-2021)...

And the last comparison is between the 5128 DF and another DF curve that I found in a recent video by Audio Precision, which I believe is also for a recent configuration of the GRAS KEMAR HATS system...

Again, I have some serious doubts about the accuracy and reliability of the GRAS DF curves used in these comparisons, in terms of their consistency with the actual headphone measurements done on those rigs. So I think all the above results should be taken with a very large grain of salt. None of the above curves are showing significant differences in the levels in the lower frequencies between the 5128 and GRAS though. The 5128 DF curve does appear brighter than all three GRAS DF curves in at least some spots in the treble though, particularly around 7 or 8 kHz (or so it would seem anyway).

I highly doubt that all those differences are completely correct though. Because most of the GRAS DF compensation curves used in the above comparisons seem a bit too depressed to me in portions of the lower treble when compared with the actual headphone measurements done on those rigs.

I think the GRAS DF curves (at least the first two anyway) are possibly also a bit too bright to be totally correct in the upper mids as well. And suspect that's why the first two in particular are making the 5128 DF seem more depressed in the 2k range by comparison. The third curve looks like it might (again, might) be a bit more accurate in that area though.

There probably are some differences in the way the two rigs actually measure in the treble, and maybe also in the bass, but I sincerely doubt that the above graphs are giving us a completely accurate idea of what those differences might be. So I wouldn't recommend using any of the above as (for example) a correction curve on the current Harman target.

The 5128 DF curve, by contrast, looks like a better approximation to me of how that rig might actually perform and measure with a spectrally flat diffuse sound field in the higher frequencies. Based on how it compares to some actual headphone measurements on that rig. And it certainly appears that the folks at HBK made quite a concerted effort to ensure its accuracy. (Perhaps because they knew that was likely to be one of the few good reference response curves that users might have for comparison with their own measurements on the rig, in the short term.)

 

[ADUHF]

A couple plots of one of my early (pre-beta) 5128 targets. This curve is just the 5128 diffuse field curve combined with a -1.25 dB per octave slope to better approximate the darker tilt of neutral loudspeakers in a room. The finished version of this target will probably be about 1.5 dB higher at 63 Hz in the bass, and also higher at 6.3 kHz in the treble. And lower by roughly the same amount at around 2 kHz in the upper mids, to approximate the average sound power response of neutral loudspeakers in a room a little better. The vertical steps on the graph are 5 dB...

Difference between the above raw 5128 target and Harman's raw 2018 over-ear target. The vertical steps on this graph are 1 dB...

Again, bass will go up a bit more at around 63 Hz on the final version of the 5128 target. So their won't be as much of a dip there as currently shown above. There are some pretty good neutral loudspeakers which have a very linear sound power response that will approximate a slope in the -1.0 to -1.5 dB per octave range though. Which is what the above target would be designed to approximate.

Some examples...

https://pierreaubert.github.io/spinorama/KEF Reference 5/KEF/index_vendor.html
https://pierreaubert.github.io/spinorama/Selah Audio Purezza/ErinsAudioCorner/index_eac-eq.html
https://pierreaubert.github.io/spinorama/Mesanovic RTM10/Mesanovic/index_vendor.html
https://pierreaubert.github.io/spinorama/Kali IN-8v2/ErinsAudioCorner/index_eac.html

 

[ADUHF]

More graphs.

A little complicated to explain, but this is the sort of data I've been studying, to try figure out how to tweak a slope to more accurately approximate the sound power response of a typical loudspeaker with a neutral response.

As mentioned in another post above, the estimated sound power responses of neutral loudspeakers more or less approximate a slope in the -1.0 to -1.5 dB per octave range over a good portion of their frequency ranges. What I've actually done in the 5 plots above though is apply a +1.26 dB per octave correction, to remove that downward tilt or slope you'd normally see in a sound power curve. To try to get a little better idea how speakers can vary in their SP response from a constant slope.

The curves on the above plots actually represent the average responses for several different groupings of loudspeakers, as follows...

Description:	Graph Amplitude Range:
10 "V-shaped" loudspeakers with a flat direct response	4 dB
25 loudspeakers with a flat direct response	2.6 dB
Top 10 best extended loudspeakers with a flat direct response	3 dB
Top 5 best extended loudspeakers with a flat direct response	4 dB
Top 3 best extended loudspeakers with a flat direct response	6.67 dB

The dB scales/units work a bit different on each graph. And don't always align perfectly with the grids that are shown. So I've listed the total amplitude range for each graph in the column on the right.

When you begin to look at the average sound power responses of multiple speakers together, the amount of variation from a constant slope actually starts to become rather small. And is generally only on the order of +/-1 or 2 dB across most of the frequency range (except in the sub-bass where the response drops off much more steeply). This is why the amplitude ranges on the above graphs are so comparatively small.

There are some definite patterns that start to become apparent though when you look at sound power data in this way. The midrange driver frequencies will tend to be flatter (generally speaking) than the lower and higher frequencies. And will adhere a little better to a constant slope between about 200 Hz and 2.5 kHz, except at the cross-over of the midrange and tweeter drivers at around 2 kHz, or so. The 10 V-shaped speakers in the first plot above were hand-picked specifically for this cross-over trait. So that's why they have such an obvious "V" in their response at 2 kHz. (The depth of that V-shape is only about -1.5 to - 2 dB below the average level in the mids though.)

Higher-end speakers will tend to have less of a dip there, because of their better cross-over designs. As the sample size increases though (from 3 well-extended high-end speakers to 25 speakers that vary more in quality), so does that dip at around 2k. Because the lower cost speakers in the sampling don't generally have as good cross-overs as the more expensive full-range models, with more drivers.

The other features that stand out are the bumps in the treble and bass. Which are (imho) somewhat symmetrical in height and shape on better-extended speakers (see the top 5 best extended speakers above). This is one of the main reasons why I wanted to look at this type of data, to get a better idea of the shape and amplitude of those humps. And also that dip at 2k in the speakers' cross-overs... So I can better apply those features to my target response curves for headphones.

 

[MayaTlab]

I'll offer some detailed thoughts when I'm back in the office. (I'm on a much-needed vacation.)

Thanks for the presentation. I’d love to hear more thoughts indeed.

The individual HP measurements in Sean Olive’s presentation showed some interesting behaviours, some of which you may have comments about :

Below 1kHz :

• A number of closed backs (K550, DT770pro, N700NC - measured passively ? -, PM3, etc.) - but not all (Bose for example, even though the AE2 doesn’t have ANC I believe and the QC35II was measured passively) - showed a sharp drop in bass response on the 5128, possibly because of sealing issues, but the K701 did as well to a noticeable degree despite being a fairly open design.
• Other open backs showed a slightly brighter “tilt” of their response below 1kHz (Utopia, HD650), but I’m a little bit skeptical that seal was an issue in that case.
• Below 150Hz or so, some HPs exhibited a number of more or less significant high-Q dips that are reminiscent of what Solderdude termed “pad bounce” in this article : https://diyaudioheaven.wordpress.com/headphones/earpads/, which seemed to shift with pad compression, and are sometimes not appearing when performing real head measurements (at least for open HPs like the HD650). I can’t help but notice that for most of these HPs the frequency of these dips shifted between 5128 and GRAS.

In the 1-5kHz range or so :

• some HPs produced a FR curve that is somewhat similarly shaped to the GRAS rigs (HD650), seemingly unaffected by quite a few variables (you could normalise the Gras and 5128 curves at 2kHz and they would overlap for the most part in that range), while others showed a significantly differently shaped FR (QC35II for example - I’m not surprised about this one in the slightest, even measured passively, but would love to have your comments on that).
• The K550 seemed to experience that throughout a wider range, down to 300Hz. The shape of the two curves is noticeably different.

5-10kHz :

• Well, it’s different, and it varies a lot. Expected. Are there individual headphones that are closer to the calculated average ? Are there characteristics to a pair of HP's design that would enable them to behave closer to the average than others ?

 

[ADUHF]

Some great observations (and questions), Maya. Thank you.

 

[ADUHF]

I'm doin my best to hold my tongue on the bass difference thingy, btw, until some others with more experience with either of the two rigs can offer comment. But I also found these remarks by Dr. Olive in a related topic on another forum pretty interesting as well...

The thing is, at least these fixtures made by GRAS and B&K DO meet an IEC standard. The 5128 is most different from the IEC 711 couplers and there is a new standard to include it. Beside the HF differences which could be pinnae related is the low frequency response. Some of this could be related to the larger volume of the 5128 coupler. The other difference is how the headphone seals to the shape of the head/cheek which always creates a greater leak than the flat plate of the GRAS 45 CA. We see similar leaks with the KEMAR head compare to the GRAS45CA. The question is which fixture best simulates average headphone leakage measured on humans?

One more thing: My intention of this presentation was not to disparage the B&K 5128, and I don't think this presentation did that.

We own a 5128 and I hope to use it more in our research. The problem up until now was our research was based on a different test fixtures, and we needed some way to be able to better interpret headphone measurements made on the 5128. That was the motivation of the research.

Jude maintains that the 5128 is more accurate than the other test fixtures. In terms of representing the acoustics of average human ear canals that may well be true. But there still needs to be some way of interpreting headphone measurements made on the 5128 in terms of human perception of sound quality. In my view, it doesn't matter if the 5128 is more accurate, if there isn't any data on how the measurements correlate with listeners' sound quality ratings.

So until someone has done a similar study of subjective and objective headphone measurements made on the B&K 5128, this Harman Target customized for the 5128 is the best solution we have for interpreting what the measurements mean in terms of how the headphones sound.

I can also completely understand and sympathize with why Dr. Olive and Harman would not want to simply throw all of their previous work and research with the GRAS rig out, and just start over again completely from scratch with a new rig like the 5128. Especially if they have some concerns about its potential reliability or accuracy. However, I still think there are some possibly much better ways of deriving a target for the 5128 than simply porting over the software from the previous rig.

There are obviously potential upsides and downsides to either approach. But I prefer to let the data from the older rig go at this point. And start with a clean slate on the 5128 instead. Even if it may still have a few bugs or quirks that might still need to be worked out. (Which would generally be true of almost any new piece of hardware. But still remains to be seen and confirmed in this case imho.)

I'll also reiterate again that I wish HBK would post some videos of these conferences, so the rest of us who were not able to participate could give them a somewhat better look and listen.

 

[ADUHF]

Description:	Graph Amplitude Range:
10 "V-shaped" loudspeakers with a flat direct response	4 dB
25 loudspeakers with a flat direct response	2.6 dB
Top 10 best extended loudspeakers with a flat direct response	3 dB
Top 5 best extended loudspeakers with a flat direct response	4 dB
Top 3 best extended loudspeakers with a flat direct response	6.67 dB

Just a quick followup to show how one might derive a 5128 target response curve from some of the above sound power info. This is my rather crude attempt at approximating the offsets to a slope, illustrated in the sound power data shown in my previous post above. The only slight adjustment that I made was to extend the falloff in the bass all the way down to 20 Hz (mostly for neatness)...

I also chose to make the boost in the bass and treble the same height or amplitude, based on the responses of some of the better-extended speakers shown in the previous graph above. Rather than using the larger samplings (with less well-extended responses) as a guide for the response in that area. This is just a rough preliminary guess though at an appropriate correction curve.

Combining the above offsets with a -1.26 dB per octave slope, and the HBK 5128 diffuse field compensation curve results in a (potential) target response approximation curve that looks something like this...

The best match I have seen so far to this is an early version of the SoundGuys 5128 target. Which I believe was based just on averaging together the responses of a number of different headphones measured on the 5128 rig that they felt were of reference quality. The overall shape of the two curves seems very similar to my eyes. But YMMV on that.

The example 5128 target response curve shown above is still very much a work in progress btw. Which hasn't been tested with any actual headphone gear yet. And like Harman's target for the GRAS system, it would only be useful as a general approximation curve. Because it doesn't capture enough fine detail in the treble, to really be useful for more accurate plotting in that range. For those who are still somewhat new to headphone graphs and plotting, it's important to understand that there really should be some bumps and dips in a headphone's treble/high frequency response (particularly in between the peaks at around 3k, 8k and about 16k, in this case). And that is completely normal behavior for a headphone, which should not generally be smoothed over or corrected out!... Just as with the GRAS FR data. The peaks in the headphone's treble response should probably follow a curve something like the one posted above though, for something approaching a well-extended neutral response.

The main reason for posting this here is simply to illustrate that there are, in fact, other ways of deriving a target response curve for a rig like the HBK 5128, based on science and high-quality measurements, which may not necessarily have to involve alot of expensive subjective listening tests. (Though it would certainly be nice to eventually have some of that done as well. As well as some actual in-ear measurements of higher quality neutral loudspeakers in a typical domestic listening space, made on the 5128 HATS rigs.) Using HBK's excellent 5128 diffuse field response curve, combined with some kind of a slope that roughly approximates a speaker's sound power response, is not such a bad place to start for something like that, imho, if you have no other resources at your disposal. And one of the rather nice things about using an approach like this is that you can be somewhat selective, and experiment with some different types of speaker response curves to use as your target than the ones I've used for the example above.

Maybe you don't want a response curve that will extend all the way down into the sub-bass frequencies like the Harman curve, or the above 5128 plot, for example. And prefer to simulate something closer to a bookshelf speaker, or something like that. Or prefer to approximate a more linear in-room response, perhaps a bit more like the KEF Reference 5, or Kali IN-8v2 in one of my previous examples above. That is something you can also easily do using speaker sound power curves and HBK's DF curve as a general guide, if you so desire. The control on what to do on something like that is completely in your hands... if you want it to be.

This is the difference between the diffuse field-derived 5128 target shown above, and Harman's 2018 over-ear target for the GRAS rig, btw, in case anyone's interested in that...

 

[Tonmeister2008]

Thank you for the above reply, Dr. Olive.

I'm still looking over some of the info that you've posted here and on your Twitter account. But should have a few thoughts on the above before the weekend is out.

Re the differences in bass measurements between the 5128 and GRAS rigs, it would help if you could post some comparative FR data for some of the other headphones that were used for computing the correction curve for your new 5128 target. So that I and others could see a little better whether the responses on the DCA Stealth and AKG K701 shown above are fairly typical of the differences. Or in the more extreme column... if that's something that you can share. If you're not ready or prepared to put any more info out on this though, then I'd certainly understand and respect that.

Just to be clear, I have no interest personally in trying to extrapolate a 5128 target response curve from any of the above data. I'm actually very pleased with the results I've been getting using sound power and the 5128 diffuse field compensation curve for that. (And may also post a little more on that later.)

I'd also be curious to hear some of Jude's thoughts or impressions on the differences in response between the rigs, since he apparently encountered some similar discrepancies on some of his IEM measurements. I believe those turned out to be related more to differences in the design of the simulated ear canals though, rather than being related to poor coupling between the headphone and rig (ie leakage or seal problems). I think this video (which I suspect you've already seen, though maybe some others here haven't) explains a bit more about that...

I watched this video again. I do not dispute that the new B&K 5128 may be more accurate than the old 711 at low and high frequencies. That is not at issue.

However, the measurements of different samples of an IEM headphones on different 711 couplers, measured by different people at different insertion depths (we don't know what the depths are ) represents a lot of uncontrolled variables that make it difficult to say what is the exact cause of these measurements differences. Why not do a more controlled comparison?

I understand the issue with different coupler volumes between simulators and how headphone insertion depth can change the effective volume.

It would have been nice if B&K had actually shown the differences in measurements between ear simulators using an Etymotic ER rather than confirm the volume it sees is different.

We purchased one of the first 5128's and I've been to their labs in Denmark, and I've also been to GRAS labs. I have no personal stake in either company succeeding. However, since our research was based on a GRAS it is the only reference we have for correlating subjective measurements with objective ones, and until a similar body of research exists for 5128's we are "stuck" with what we have.

The presentation I gave focussed on measurement differences between different test fixtures for AE/OE headphones, and the differences seem largely related to differences related to pinna up to 8-10kHz and coupler (above 10kHz). At low frequencies the differences are largely attributed to seal, which I suspect has nothing to do with the couplers, and everything to do with the pinna/head/cheek interface with the headphone. Heads generally provide less reliable seal than a flat plate. And softer, more pliable pinna also give better seals. You could argue that one head is more anthropomorphically accurate than the other and the leaks we see are therefore more representative of humans. But where is the research and data to support that?

With IE headphones, the differences between test fixtures will likely come down to the couplers, and if you use a pinna, how easy it is to get a reliable seal. Rtings decided to not get a 5128 because they could not get a consistent IEM seal/measurement compared to a GRAS45CA or their own Head Acoustics Fixture.

 

[ADUHF]

Thank you for the follow up on the above Dr. Olive.

I will have some further thoughts on this. But also wanted to give folks here the heads up that some of the presentations from this conference are now available on HBK's YouTube channel...

https://www.youtube.com/channel/UCOUjyYbxfOc14yuEMG796VQ/videos

Including...

Erik Ziegler's - Keynote Presentation
Martin Alexander's - ElectroAcoustic Principles – The Complete Idiot’s Guide to Basic Electroacoustic Measurement
Mark Serridge's - Why do so many of us like the sound of analogue recordings on vinyl when the measurements look so bad?
Søren Jønsson's - Wideband impedance measurement technique in the human ear canal
Lars Birger Nielsen's - 5128 High Freq HATS and conformance to the new ITU-T P57/P58 standard

It doesn't appear that they have posted Dr. Olive's Modeling And Predicting Listeners’ Headphone Preference Ratings or Jude's Audio Measurements As Consumer Content though yet. These have been coming out a little at a time though, for the last few days. So hopefully these other presentations will also be posted soon.

 

[ADUHF]

Just wanted to give a little heads up that communication may be a little spotty from me on this and other topics here on Head-Fi for awhile. Because I'm in the process of moving at the moment. I have enjoyed reading some of Dr Olive's and Maya's insights though on this subject. And hope maybe to get back into some of the above a little more when I'm either a bit more settled, or have a little spare time to think a bit more on some of this. I also appreciate some of y'all hearing me out on a some of the above as well. And hope to get back to this soon.

 

[ADUHF]

It doesn't appear that they have posted Dr. Olive's Modeling And Predicting Listeners’ Headphone Preference Ratings or Jude's Audio Measurements As Consumer Content though yet. These have been coming out a little at a time though, for the last few days. So hopefully these other presentations will also be posted soon.

As of this morning, both of the above videos are now also available on HBK's YT channel. So here they are! (yeah)

 

[ADUHF]