“All groups preferred a significantly higher amount of
treble than specified in the Harman curve. People with lit-
tle hearing loss wanted about 12 dB more level at 16 kHz.
Each group wanted progressively more treble boost to com-
pensate for increasing hearing loss, approximating 3 dB for
each 10 years of age over the age of 30. Within any group,
the treble boost could be adjusted ±3 dB (one step size
in the test material) with only a small drop in the group’s
average quality rating.
Fig. 14 shows the frequency response curves that each
group preferred. Although the preferred curves are much
higher above 10 kHz than the Harman curve, the levels are
not so high as to be impractical to obtain with reasonable
hardware. The preferred level at 16 kHz for the youngest
group is similar to the level at 1 kHz. The preferred 16-kHz
level for the oldest cohort is similar in level to that of the
3-kHz peak.”
Any thoughts on their methodology, data, conclusion?
Haven’t read the study yet. But it would not surprise me if people generally prefer a little more high frequency info than the Harman curve. I can’t comment on the in-ear target, but have always felt that the Harman over-ear target under-estimated a neutral response in the higher frequencies.
High frequency hearing loss is also well-documented in older people. So again, a preference for more HF wouldn’t be very surprising there. I am not an ear doctor or audiologist, but I would have some concerns about using headphones as sort of a hearing aid to try to compensate for this HF loss, due to the possiblity that it could accelerate the loss further. And I try to stay pretty close to a neutral response in my headphones and other gear even though I’m over 60, and most definitely have some HF loss.
The level at which you listen to music and audio content could also effect your preferences in both the higher and lower frequencies due to the Fletcher-Munson effect. So one other possible conclusion to this study could be that older people prefer to listen at lower volumes, and might need some more HF to compensate for that. Maybe the listening levels were taken into account in the study though. Since I haven’t read it yet, I’m just spitballing on the conclusions, and some of the possible reasons that people or groups might prefer a bit more HF than Harman or other groups.
Some older people might possibly also prefer a somewhat brighter sound in some cases due to “hidden hearing loss” or the “cocktail party effect”, which is discussed a bit by Floyd Toole here…
Different measurement rigs can also measure differently, especially in the HF. So if they were not using gear identical to what was used in the Harman research, that might also help to explain some of the discrepancies in the HF. I have not looked at the details in the study though to see if this was the case.
There was also a study done by Senselab not too long ago which seemed to suggest a preference for headphones with brighter treble/HF than Harman. I do not know what the current status of that study is though, and whether any of its conclusions were ever updated, revisited, or re-verified. But this was even after taking into account some of the differences in brightness between the GRAS rigs used by Harman for their research, and the HBK 5128 which tends to measure a little more brightly in the treble/HF, and was used for measurements in the new study.
The study was discussed by Doctor Olive at a recent CanJam presentation shown in this video, which is pre-cued to the Senselab section. Based on looking at many other headphone measurements on the HBK 5128 though, I think it is quite possible that the “best rated curves” shown in this Senselab study over-estimate a neutral response in the treble/HF.
Small footnote: Despite some of the comments in the video, the Harman target shown in the Senselab study slide appears to be Harman’s usual 2018 over-ear target for the GRAS systems, rather than some kind of translation of same for the HBK 5128. It looks slightly different though because the Senselab graph only goes up to 16 kHz, and the top 1/3-octave of the Harman curve is missing.
In the most recent Harman IEM listening tests (presented at CanJam SoCal), Dr. Olive identified something many IEM enthusiasts, including us, have been saying for a long time. That it’s not about a single preferred target curve, rather it’s about proportional bass and treble, and that’s ultimately what we should be shooting for.
For example, the highest scoring result, the SoundGuys curve, which is similar to to JM-1 + filters, was statistically tied with Harman IE 2019. And it’s not just that it was similarly preferred across a wide range of listeners, it’s that the same people rated both curves similarly. And to anyone who knows what these curves look like, that should be surprising, given just how different they are.
The key unifying element is that the bass to treble delta is similar between those two curves, even if they get there very differently.
Sean also indicated that it may be worth adjusting Harman IE to have more upper treble and less lower/mid treble, which is also something many of us have been suggesting for some time. While my compatriots are probably less partisan on this issue than I am, it is some vindication for those of us who had criticized Harman IE 2019 in the past.
“Any thoughts on their methodology, data, conclusion?”
Methodology
Seems like well-done science, though making amendments to Harman’s target while using a quite different replicator IEM makes me think this paper’s findings may be marred by unknown differences in HpTF in an area we already know to be the worst-case scenario for IEM-to-IEM uncertainty.
Data
Seems pretty uncontroversial that a target with so much lower treble would be better balanced by more upper treble. Also seems uncontroversial that older subjects would prefer a brighter sound.
Conclusion
Don’t really think its worth adopting this as a standard of any sort, as opposed to letting research done on a more accurate ear simulator lead the way forward in that respect… or better yet, let studies based on measurements in real ears lead the way.
This was exactly my thought as well. Balancing out what many would consider… problematically forward ear gain with more upper treble seems like it would be generally more pleasant, despite the overall level.
If you don’t mind could you post those curves here side by side? I looked them up but didn’t want to add confusion to the thread in case I got the wrong curves. I wanna see how those curves compare to the AES paper’s proposed changes.
Pink is the SoundGuys target, while blue is our chosen DF HRTF for IEMs—JM-1—with the filter adjustments from Harman’s 2018 target applied.
We believe that the blue curve and the pink curve would test similarly in listening tests due to being virtually indistinguishable below 8 kHz.
To be clear, the blue line is a set of well-testing preference filters applied to a DF HRTF we use—it is NOT a target that we are offering/prescribing to people as the ideal target for IEMs. It just kind of shows that our chosen HRTF measures very similarly to a well-performing target (SoundGuys) when popular/well-performing over-ear headphone adjustments are added to it.
Below is the difference curve between SoundGuys and JM-1 + Harman AE/OE 2018 filters:
Less than ±.75 dB difference for practically the entirety of the response except for the last octave.
And here are the above two compared to Harman IE 2019. Personally I think a little bit more upper treble than Harman is likely an uncontroversial change, but perhaps not as much as SoundGuys.
Thanks for all the graphs! The sound guys target has that little rise at the upper treble, similar to the one in the research paper for the forum topic?
I vaguely recall hearing on one of your videos or podcast that you guys didn’t like the method of how Sounds guys arrived at their target, even though the target itself is rather well liked. Do you know how they came to the idea of extra upper treble energy?
You may want to include IE 2019 in there, just because Sean did mention (even if jokingly) “we only use the good one”, indicating not 2017 since its ear gain was the strangest.
SoundGuys’s target is based on an average of “good” over-ear headphone measured on the 5128, as well as apparently being somewhat weighted by the DF/FF response of the 5128 as well. To be clear, they have not disclosed the exact calculation, just a rough outline of the method.
The reason I myself am not a fan of this is because the 5128 in particular seems to be the “brightest” headphone measurement fixture when it comes to its average HpTF for OE headphones relative to the IEC 60318-7 heads (GRAS 43AG/45CA, B&K 4128, HeadAcoustics HMS II.3).
Below is an average of the same models (but not the same exact units) of 14 different headphones measured on each HATS fixture—5128 in pink, GRAS in orange, 4128 in green, and HMS II.3 in blue. To me, it’s fairly clear that the 5128’s measurements of OE headphones are the brightest above 10 kHz or so.
Over-ear headphones are different than IEMs though, because they interact with a portion of the anatomy that the former doesn’t—the outer ears. This is why we don’t use the 5128’s DF HRTF for IEMs and instead use JM-1—the 5128’s outer ear interactions are unlikely to be as close to an average human as we’d like.
For those reasons, I don’t see the point in baking in outer ear features from headphone measurements that might not track with how IEMs interact with the same rig.
Especially because you can actually see even with the simple average of 14 headphones in my measurement that the rig itself is engendering the characteristics in the area I’m talking about. Of course, you can see its brighter than the other heads, but this average HpTF for 14 headphones—which are likely different than the ones used by SoundGuys—also has the same small 12 kHz dip followed by a rise present in the SoundGuys target.
And note, this is not the delta for non-HRTF related HpTF variation, as observing trends in how headphone behavior is impacted by different rigs is a different yet perhaps even more intriguing datapoint. This is in reference to HpTF being a headphone’s response on a given head.
So for folks unaware, the HRTF for each of these is different as well, and importantly the pinna effects are also brighter on the 5128 compared to the other rigs/ears, and we expect brighter than the average human’s pinna effects. This is the reason to use non-specific pinna effects based around the population average response for IEMs, given they bypass the pinna and need to assume those features.
Methodologically dubious because it effectively entails a commitment to the idea that “headphones perform this way, therefore they should perform this way”, which is a very well known fallacy.
Yeah, doesn’t make much sense to have a target based on what manufacturers are making and then averaging it out. It makes a little more sense if (I assume they did this?) they average out the curve of only the best sounding ones. But it’s still not as rigorous of a method as what Sean did.
Still, interesting to note that they coincidentally got that treble hump using a pretty round-about way compared to this research paper’s method.
When/if/how would you guys update the target you use for reviews? And also what are your thoughts on the OE target’s upper treble curve? Should that be modified? What are your personal preferences regardless of the curve?
Well we don’t use a target. The whole concept of a target curve is somewhat antithetical to the outcomes of the preference research, which is that preferences fall within a range. So that’s why we use the preference bounds. As far as what’s likely to change, we may update the bounds to be narrower in the mids, but wider in certain places in the treble, since that seems to be more congruent with the research.
But also, we want to start showing this as shadings or a gradient, something to indicate that not everything within the bounds is similarly likely to be preferred.
The 5128 definitely measures brighter than the 711-based rigs, at least on most over-ears. I’m less sure about IEMs, since I’ve only looked at a handful of those graphs. I’m not sure if that brightness is due to the more advanced and anthropomorphic ear canal/coupler on the 5128, or something else. (Resolve also mentioned the pinna above.)
Assuming all four heads and their pinnae are close to an average human’s though (which could be assuming alot!), I’d be more apt to trust measurements made on the rig with the newer canal design. And probably use the same target (or pref curve, if you like) for both the in-ear and over-ear transducers, with the important caveat that there’s likely to be more individual variation/deviation in terms of accuracy and performance with the in-ears, due to their lesser interaction with the rig’s head and ears.
If the rig is truly a good representation of an average human head, ears, ear canal, and ear drum, then an IEM user should be able to figure out how they deviate from this norm through a bit of trial and error. And figure out what type of IEM and FR is a good fit for their particular head, ears, and ear canal… or maybe not!
With IEMs, alot of this just seems like a crapshoot. So I don’t envy those trying to thread the needle on this. It’s difficult enough to find good sound on over-ears imo. Maybe there’s a way to scan your head though, and figure out what your ideal IEM signature is from that.
Semi-important footnote: This could all be terrible advice and analysis since I don’t use IEMs at all.
It’s one of those things that are theoretically plausible but practically impossible I’m guessing. For most of the FR up to 5khz people mostly hear similarly, above that however… you’ll have to 3d scan the ear canal, then calculate the precise resonance and wave cancellation frequencies happening between the driver of the specific iem, the ear canal and the ear drum, then try to match that to a measurement rig’s data, somehow without its own wave cancellations inside the rig.
Practically speaking the treble range is a bit of a crap shoot for iems. For people who insist on no eq there’s not much they could do other than to tip roll to change the resonance frequencies by changing the distance from the ear drum to the driver, or just try out different iems until they find one that sounds right.
But if you eq then it’s possible to fix some of the issues, but it’s time consuming and you can’t fix wave cancellation dips anyway.
Not sure I agree, Resolve. It may seem dubious because we don’t know exactly how it was done, and how rigorous the sampling of headphones was. And the results may not precisely fit with your or my conception of the ideal response.
If you’re tryin to get a better idea of the characteristics of a certain class or category of devices though (such as neutral headphones or loudspeakers), then computing an average seems like a pretty logical and appropriate way to do it.
Dr. Olive also used the median (which is different than the average, but a similar idea, see footnote below for more) of the best rated curves in his Senselab presentation, for example. And your homey also just used averages of 14 cans in his graph above to illustrate the difference in HF response between 4 different rigs. So this kind of thing has its place imo.
This sounds suspiciously like averaging as well…
(Footnote: The Senselab slide says “mean of the blue region”, but the trace appears to be the median or middle of the high and low extents of that region.)
Since you guys have the equipment to measure , have you done experiments to EQ completely different headphones ( different drivers , geometry, etc) and done blind tests to see if people can tell the difference ?
That would be pretty interesting.
Of course someone experienced may be able to tell the headphones just from their fit.
But nevertheless, I would be super interested to see if most of everything is FR , that would be a good way to test it.
On the flip side it wouldn’t be the best for sales of different headphones. But then again technical aspects such as distortion would be important for headphones to have so they could ‘ hold up well to EQ’
