Developing a new headphone reference target

I wanted to provide some commentary for our video on moving away from the Harman Target as we start using the new B&K 5128 headphone measurement system:

I’ve posted this on Head-fi as well, as there’s been a bunch of discussion surrounding this topic, and I felt it was important to also have that here to get some discussion and feedback from folks here as well - so this is a bit of a crosspost.

There are, in my view, two ‘good’ ways of developing a target curve for the 5128 that achieves a result people actually like - and that’s important because theoretically you could just use DF for everything and then represent it against a slope, but I tend to think that’s more confusing (and this part is just my opinion, I’m not suggesting it for the others who are helping with this target development).

1. One is to calculate the Harman in-room response for the 5128 (and there are ways of doing this), and then apply the shelf filters from the research. So in other words, get to the same starting point for the new system, and then use the results of the Harman research to specify the shelves. This approach would result in a target that looks somewhat similar to the existing Harman target, with the bass shelf, but is then applicable to the new standard. This is also the approach I would expect Dr. Olive to take if they were to fully develop a public target. At the moment though I think they’re just using a stop-gap internally based on the delta of results, and unfortunately that doesn’t quite work for a reference curve since headphones behave differently depending on the head they’re on - not to throw shade at them or anything, it’s understandable as I don’t think the 5128 is the focus of current research. I suppose if they were able to validate the same on-head behavior among the different systems somehow it would justify continuing with that, but that’s not a study I’d expect from them anytime soon.

2. The second way is to rely on diffuse field head-related transfer functions (or DFHRTF). In overly simplistic terms, think of this as like… a flat speaker ‘calibration’ that each measurement rig has (and every head/ear), and conveniently it’s also very close to the Harman in-room baseline. So while DF for the GRAS doesn’t look like DF for the 5128, compensating to them for their respective rigs yields a comparable result. But because people generally don’t like the way DF sounds (it’s too bright for most listeners), we can apply what the Harman research validates - that people prefer a downwards tilting slope for both speakers and headphones - to the DFHRTF of the 5128. There are a number of benefits to this method, like that you can more easily get a ‘high res’ version of the target, and on the practical end, you can apply the same concept for comparable results to any measurement fixture or even in-ear mics on real people. Note - they won’t necessarily look the same, because headphones behave differently depending on the head/rig. But they will be using the same target principle, and thus comparable.

So, in my view both of these approaches are valid, we just think there are more benefits with the latter, and that it fits more closely with the wider body of acoustic research than just Harman. Also… and this is a point I’m not sure how much I care about… but going with the DF plus slope approach also means we’re not limited to Harman conditions - meaning the specific speakers/room used for that research. Going with the former approach would mean trying to… ‘inject’ previously baked-in anchors that aren’t really needed. I think the counter argument here is that we’ve been assuming those conditions for years so we might as well go with what’s familiar. But this also means future outcomes are slightly ‘poisoned’ by those conditions as well - I say in scare quotes there because as Dr. Olive has pointed out recently, you could do a lot worse than the Revels that were used.

One last thing, I just want to anticipate commentary on the 5128 not being meaningfully better than the older GRAS systems and some of the thoughts around chasing the new measurement standard turning into gatekeeping. While it’s true that the older rigs are still useful, as Jude has pointed out a number of times the B&K 5128 is a better representation of an actual human than ever before. In practice, it’s not more ‘accurate’ for where the majority of audio information falls (between 200hz and 10khz) but below and above that, it is better. How much that really matters is still up for debate since at high frequencies positional variance is a massive factor anyway - not just on rigs but on real people too. But review platforms getting these new rigs isn’t just about benefits in the minutia, but rather moving to the the new measurement standard the 5128 has pioneered.

Then, ideally, with our proposed target approach we can make even the inexpensive clone rigs comparable up to a certain point. At the moment that’s not doable because the shore hardness of those pinnae is too stiff, meaning you can’t actually use the GRAS DF for it. But I’m told there is a closer pinna to the KB5000 in the works for those as well, which could in theory make our proposed target concept applicable. In my testing when I used an official KB5000 ear on the clone coupler, there’s still a resonance at around 12khz that shows up every time, but apart from that it was quite close to the official GRAS. The bottom line is that our intention with the DF plus slope approach is to provide a target that makes more measurement rigs cross compatible and have the opposite effect of gatekeeping older or less expensive ones - even if there is still some work to be done there.

Anyway, we’ll be doing more videos on this topic soon - maybe like an FAQ to address some concerns/questions. But I figure I can also do a bit of that here, just keep in mind that our target is still a work in progress:

EDIT: Here’s the FAQ Video

Q: Does this mean the Harman Target is bad or otherwise unimportant?
A: No, we need to distinguish between the specific target many of us are used to with measurements done on GRAS KEMAR based systems, and the large body of research behind it. We want to make use of the research outcomes, like the bass to treble delta that people preferred, but can’t port their specific target that reflects that over to the new system for the reasons mentioned above. Also, the Harman Target is still great for GRAS systems, and we will still be able to use it as a reference point on those systems.

Q: Will we be re-measuring all of the headphones that had previously been measured on the older systems?
A: Yes and no. We will of course be re-measuring many headphones on the new system and providing those graphs, but we don’t have access to everything in the back catalogue. We can, however, apply the new target to that data for cross compatibility. For new headphones that come in we’ll also be measuring them on the GRAS as well to see how they vary across different heads.

Q: Why not just show raw graphs for everything?
A: While we’ll continue to provide raw graphs, those done on GRAS systems will look different from the ones done on the 5128 (or any other system), and so there has to be a rig-specific reference target associated with the data. Additionally, as a matter of what defaults should be, compensated data is technically better because it avoids the common illusion of reading parallel lines against one another. There will be a video on this, but in short, yes, we intend to provide both.

Q: Why not just use a well-known headphone like the Sennheiser HD 650 as a reference point?
A: While this is useful as a point of comparison - and I imagine would be useful to many who have heard that headphone - this approach won’t work for anyone who hasn’t heard that headphone. Additionally, this is less helpful for evaluative purposes, because even the HD 650 has a particular ‘flavor’ to it - as do all headphones.

Q: Can’t you just EQ a headphone to Harman on the GRAS and then use the 5128 result as the reference point?
A: No, largely because headphones behave differently on different heads, as mentioned earlier. But also even if a 1:1 match were somehow validated, this would still default a heavily smoothed outcome and we’d lose the ‘high res’ option.

Q: Will you have a measurement database to compare results?
A: At some point, but until then published results will be compiled in a forum thread like this one.

And of course, there will be more to come as we fill in more of the picture with this new target concept.

11 Likes

That’s certainly enough for me to say “makes sense - works for me.”

The fact that you can create a more high resolution target that has the potential to be useful for the “older” rigs sounds like a no brainer to me. We trust you and your team (we know that y’all have the qualifications to do this), and I know I’m very excited to see the final outcome.

Do you plan to “pilot” the target with a selected group of listeners when it’s “ready?”

5 Likes

Not publicly at the moment, but we’ve been having discussions around how we each perceive headphones comp’d to the new target. The thing is, we can also make changes if we need to. There are still questions surrounding what the slope degree is and where it starts (going with 20hz at the moment). I think so far the only issue has been what happens at 2khz with DF as results tend to show a bit of a dip there with a number of headphones. But it might just be that we’re used to a dip there and in practice we should be EQing that up more.

2 Likes

I can imagine new EQ presets coming for headphones that measure considerably different on the new rig, no?

It’s a bottomless rabbit hole, baby, and I love it! :slight_smile:

There will be some differences, but headphones that ‘measure well’ on the old system should still measure reasonably well on the new one. The key difference at the moment is the highest part of the ear gain, which is slightly shifted between Harman in-room and DF, but that’s also in part because of the 1/2 smoothing of the Harman target. So we’ll see if we stick with full res, but at the moment that’s the plan.

Right now, 2khz is the egg being broken to make the omelet.

3 Likes

Hahahaha. Love it. Good luck. Thanks for continuing to advance the industry and the hobby, man!

2 Likes

Also the higher the frequency the more the individual HRTF varies making measurements above 10k much less useful. So really the biggest advantage I see in the 5128 is increased accuracy in the lower bass region. This will be great for IEMs but I’m skeptical we’ll see much difference for headphones.

Regardless of my nit picking at any real increase in accuracy, it seems everyone is moving to the 5128 and being able to compare measurements across multiple sources is more important. I think approach 2 makes the most sense, although it sounds like the resulting target won’t have a bump below 200hz like Harman. Unfortunately for me I am very sensitive to how a headphone performs in that same region.

Bottom line I trust you to come up with a target that, given enough time, represents how a headphone truly sounds. We can’t forget that Olive was more interested in what the majority of consumers want to hear than what is true to the source.

PS I’d really like to see an online tool for EQing using the headphones.com measurements similar to what @crinacle has. Y’know… In your spare time… :grimacing:

4 Likes

Hmm that’s the same region I tend to leave in a dip compared to Harman if the headphone already has it. I see it a lot in Hifiman and Sennheiser headphones and if I EQ it up to Harman they get too shouty for me.

1 Like

YES, please! Seconded! :slight_smile:

3 Likes

And I’d be willing to pay a reasonable subscription fee for access too

2 Likes

+1 on this. Squigs are just too much fun!

3 Likes

Yes, better EQ tools are needed for sure. But I also would caution folks on EQing to targets past a certain point or applying profiles in a fire and forget manner. It’s useful, but not the be all end all, because… you don’t know what the FR is going to be like at your ear drum.

3 Likes

Yes we’ve had this discussion in another thread. I think I’ve achieved Stage 9 since then. :sunglasses:

3 Likes

Hahaha, me too. It gets so hard at that point though lol

But yeah, EQing by ear after the presence region is the best approach.

2 Likes

Hey now, people offered to pay us. None of that :wink: .

In all seriousness though, an EQ tool is something that’s come up quite a bit in past discussions among the team and I imagine at this point its not really a matter of if but a matter of when.

7 Likes

Can you share some measurements ? Well known headphones like hd650, hd800s and Arya. So we can see the differences.

Yes we will show some of those soon.

2 Likes

Some additional thoughts on the new B&K 5128 target we’re working one that I feel are worth addressing:

  1. What we’ve come up with is NOT ‘the Resolve target’. This has nothing to do with my personal preferences or a reflection of what I personally enjoy the most above all else and so on. Whenever you’re considering what the concept of a ‘reference’ should be, you should - ideally - take your own personal preferences out of it. As many of you know, I enjoy tunings that are versatile for a wide range of recordings and I listen to a lot of different styles of music, meaning that what’s ideal for me might be a bit more relaxed in certain regions, or have a bit more energy in other regions. And… if you read through the Harman research, you see that it’s precisely because people have different preferences and that no singular result is preferred by everyone that this research was needed (read the segmentation paper in particular for more on this).

  2. Headphone reference targets need to have a good conceptual foundation. In theory, we could come up with a curve that sounds good to a bunch of us and call it a day, but that’s not good enough for developing a proper reference curve. We feel that using DF as a starting point is a solid, recognizable foundation, and indeed DF is a baseline for the much larger body of audio research beyond just Harman. What we’re doing with DF, applying the downward tilting slope, is just another way of achieving the bass to treble preference tilt that the largest group of listeners enjoy, just without the shelf filters and the 1/2 octave smoothing of the Harman target. So, regardless of what the end result is for our target, using DF as the conceptual foundation is a good fit with the existing body of audio research, and what we’re doing with it is congruent with the well-tested results from Harman as well.

  3. We need to test our target to see how people hear it. Those of us who are working on it have already tested a number of headphones EQ’d to match the target on the 5128. So far, results have been generally positive, however we need to get more ‘ears on’. Even though we have a solid conceptual foundation, what we don’t have is preference data on other key regions of the audio spectrum apart from bass to treble delta, and while we can’t expect to get that as comprehensively as a fully-fledged study on this topic (maybe one day), using a virtual method with EQ profiles can at least give us an indication of whether or not we’re on the right track, or if we need to make changes to one region or another.

  4. None of what we’re doing with DF plus slope locks us in to the proposed target result. If we discover certain qualities of the B&K 5128 that are consistently different (in the same way) to actual humans tested with in-ear microphones, that might cause us to make adjustments… we can do that. This is a key aspect of testing with the new system moving forward, and as mentioned in some of our videos on the topic, we’re going to be evolving our in-ear mic testing procedures, both to validate on-rig behavior and to ensure positional matching between on-head and on-rig results.

What I’ll be doing in the near future is providing a series of EQ profiles on a number of popular headphones for the community to apply and give feedback on, so once we do that, it’ll be your opportunity to give input.

13 Likes

Happy to be a lab rat for any profile of the 6XX and LCD-X 2021! :slight_smile:

2 Likes

I think the 2013 Harman target is the most neutral above 1khz, all other curves sounds like they have elevated treble. I’ve done some experimenting with a downslope diffuse field target myself but I’ve found flat from 20hz-1khz with the 2013 Harman target above 1khz sounds the most like flat speakers in a good room. Makes sense since the 2013 target was tested with experienced listeners instead of the average Joe.

1 Like