HD600 should be on the top of the list lol
Well… I’d like to think that the exclusion of the unwashed masses in the whole ‘trained vs untrained’ thing is a bit of a misnomer, or what we might take from that should probably be scrutinized a bit.
To put it bluntly, I don’t think we should so easily put audiophiles into the ‘trained listener’ category, and if anything we’re more likely to muddy the results than contribute to a clear outcome. Moreover, the whole ‘trained listener’ thing requires quite a bit of consideration as well, as I remember that what constituted ‘trained’ was a minor sticking point when reading through the research and caused me to raise my eyebrows a bit. But, at the end of the day this is the best we have for publicly available research on this topic, so it’s what we go with. Perhaps a better parsing of listener types should be something along the lines of “those who strongly give a shit about tonality”, vs “those who only mildly give a shit or are otherwise indifferent”.
This was also the initial concept behind the Harman Combined target, in the sense that I felt it was a bit more appropriate for picky listeners when it comes to bass to treble delta than 2018. I could’ve just gone with 2013 but I know a lot of us find that to be too shouty at 3khz, myself included, and I also suspect this is one of the reasons for the treble lift in 2018, since, as you increase the treble above 3khz, the glare goes away with lower harmonics not dominating as much over upper ones. Also keep in mind that the region above 10khz is unreadable due to measurement artifacts, so it shouldn’t dip down there like it does to that extent.
There is also another element to this, which is that Harman is smoothed to 1/2 octave. And if you look at Harman in-room and unsmoothed DF, they’re fairly close there with the difference coming down to the Harman conditions, like the specific Revel speakers and that room. Overall ear gain level for an unsmoothed target would be higher around 3khz than what we use too, and I think this is a bit of an issue.
My take on this right now is that this is probably at the edge of what’s tolerable for folks. Like, depending on the ear, you may have a different tolerance for that region, but also we have to consider that for certain types of music and recordings, that high of ear gain just isn’t going to work. So either we bite the bullet on bad recordings and just say too bad for more aggressive music, or we aim to be a bit more conservative in that region for the benefit of versatility - again, just a thought. We don’t have that built into anything yet on the 5128, but it’s up for discussion.
The nice thing about using DF plus slope is that you get a better view of what the treble should be in relation to the highest part of the ear gain, which so far has yielded a massive improvement to treble timbre. But, we’re of course also discovering that this isn’t the same across different headphones. So, whatever happens we’ll need to build in boundaries around the target to reflect those differences - and also the fact that people have different heads/ears.
While a product manager in a totally unrelated field, I tried very hard to find new users that had little or no experience with my product. In your quest you are right in deprecating the trained ear of audiophile testers if you are developing your new target for non-audiophiles.
OTOH, if the target is audiophiles, they are IMAO, too opinionated to present a clear and homogenous target.
Audiophiles? Opinionated? This is the first time I’m hearing of this phenomenon.
Glad to see you guys are really putting some thought and research into this. Maybe 2 targets are in order? A “reference” sloped DF (maybe with minor tweaks) target and a “preference” modified DF target with less ear gain might do the trick. Then users can go off which ever they prefer and less compromises need to be made.
I’m in line for the Arya Stealth.
In my book Arya Stealth is close to perfect for orchestral and maybe live music but when it comes to modern pop music it’s too bright and the dip around 1.5k become noticeable. Making it too lean and put emphasis on 3-5k energy.
Is there any consensus on 100hz-1khz area ? Harman have it dipped around 200hz and slightly rising up to 1k. Some prefer totally flat. I’m curious about this region. Did you notice any effect flat vs slightly rising energy up to 1k ? Bass, ear gain and treble always focus points but this 100hz-1khz region is interesting too.
Yes, so with all apologies to Crin, it shouldn’t be flat below 1khz (we’ve chatted about this). Basically, with Harman the bottom part of the ear gain dips down, and that’s because it’s technically still part of the in-room result measured at the ear drum. Not only that, it’s also contoured similarly with DF. So yeah, we should include the bottom part of the ear gain, which is what that dip is. Now… with that said, when you apply a slope to DF, it effectively lifts that whole region, depending on the degree of the slope. But it still retains the same contour or ‘shape’ to it.
This is sort of a hidden benefit to DF plus slope instead of a shelf, in that it gives a bit more presence and ‘beef’ to fundamental tones that token that range and avoids the occasional leanness of Harman for music that doesn’t fully go down into the sub-bass, which I imagine is one of the reason’s why Crin lopped it off at 900hz in the first place. So, bottom line is that yes, the ‘shape’ should be included, but for those who find it perceptually thin, a slope will likely be preferred over a shelf. Personally, I like both styles of tuning, as a bass shelf is fun too.
Same! My LCD-X 2021 is ready for some Headphones dot com preferred slope target EQ.
I look forward to seeing the headphone show 5128 LCD-X measurement too. Jude’s shows more ear-gain compared to all 43AG measurements, so I’m curious to see if the HP.com one measures similarly.
I am certainly interested in the new targets HP dot com might suggest. After all, EQ is not the devil, albeit my 2nd choice (as I am lazy). More generally, I wonder how any target might be influenced by the input, i.e., on the music being listened to. If, at an extreme, my preferred genre is bass-free, then what happens there does not matter. And if I am a BASS-head, what happens there is all that matters, treble be damned. So what song(s) exhibit(s) the breadth to allow one to assess a particular target? (I worry this might border “dumb question” territory, so please humor me.)
I can see why folks would say that, but IMO everything matters, just depending on the person more or less in certain regions. And yeah bass level is a very strong factor in overall listener preference.
In my experience, oftentimes people don’t necessarily know what ‘good’ is or what they prefer until they hear it. It’s why you get so many posts on various places indicating ‘headphone x is the best thing ever’ - but really that’s with limited experience with other stuff. Now… That’s not to say they don’t enjoy it. Merely that they may have an even stronger preference for something else if they get a chance to hear it. Those same people may be back a few months later extolling the virtues of something else they prefer. It’s kind of what makes the hobby a rabbit hole in a way.
Or, just as likely a chance to find the flaws if they listen to it long enough. Like “Motormouth” McCaskill, a girl I knew in high school.
Thanks for the response!! I have about 88 thoughts on this, none of which are answers. My mental model is (input) → (headphone/iem transfer function) → “target” → (head/ear transfer function) → (output). Where (input) is whatever the source outputs (and there is potentially a bunch of stuff in that chain!! including EQ) and (output) is what my puny little brain receives. Accuracy is achieved by minimizing ∆ = |(output) - (input)|. Happiness is achieved by minimizing ∆ = |(output) - (imagined/desired input)|. My original question was more than a bit pedantic as I think max∆ is a function of the range of inputs allowed. OK, nerd set to off. It is a wabbit-hole, and I believe that a better model of a headphone/iem transfer function (which I don’t think are really the same, but that is for another day) than Harmon does exist in this model.
As far as your model of perception goes, I think what you’ve got there is right except you can remove the “target” layer, as that’s really satisfied by the head/ear transfer function. When we talk about a target, it’s really just meant to indicate a reference point what we think might be suitable based on that particular head/rig. So kind of like… using a head and ear to give an example of what ‘good’ might look like.
On different humans that’s also bound to vary a bit - not only will it look different, but even with person-specific normalization will there be some differences. I think overall we hear things more similarly than we do differently, but yeah there’s still some variation.
As far as happiness is concerned, that’s always a tough question but again your model is a good stab at it. I do think there’s oftentimes a difference between what people think they like and what they actually like (the Gladwellian examples come to mind), so maybe there’s more to happiness than “bass is great so I want it to have all the bass”.
When it comes to the largest audiences, there’s still an enduring trend of ‘extra bass’ marketing out there, in part because for a long time headphones kinda sucked. And in large part, smaller or lower quality transducers struggle with bass reproduction - think about the speakers in your phone for example. This is what a lot of people are coming from. So it makes sense that bass has for such a long time been a determining factor in the public consciousness of ‘sound quality’. In some ways it’s taken on an inertia of its own - to the point where even people who might not like a bass boost in practice still THINK its good, and overindex on that for their expectation of what constitutes ‘good sound quality’.
This is great work. I’ve been super busy in RL and have not really being paying attention. One of the mods at SBAF pointed out what you are doing. I don’t think it was this video, but another one where you suggested a fuzzy zone target.
Reading around the Internets, looks like you are causing some trouble now with a “Resolve Target” LOL.
Anyway, the 1db/oct downward slope (at the listening position) is what is common in studios, mix stages, etc. There are variants of this like B&K AES 1970 which are more or less flat to 200Hz before the downward slope. Also, when speakers that measure flat on-axis 3 meters away are placed appropriate listening positions in decently treated rooms (or a normal living room with carpet and furniture), we will also see this 1db/oct downward slope behavior (at listening position). This is what I consider neutral. This is what I “train” my ears on.
The target I prefer to use is naturally what studios have in place. Naturally there will be variance, if they are using JBLs, Genelecs, ATC, Adam Audio, and even down to the specific speaker models.
This is why I prefer the 1db/oct slope or B&K (same thing but less low/sub bass), but I also like what you did with providing that zone which can accompany Harmon (preference research) and perhaps others (Audeze “house” sound).
Anyway, I wanted to clarify what you are doing. Basically applying DF comp on the 5128 and then plotting against proposed targets? Tempted to join the 5128 club and follow what you are doing. Thought it would be nice if we can converge on a method. More people taking measurements using the same method will lead to more comparable and understandable results for people at large.
Yeah it also seems like there’s a lot of confusion and bad faith accusations from the usual suspects going around on this. As expected, certain people are playing the authority/narrative game rather than engaging with what we’re actually putting forward.
To clarify a bit, I would certainly not call this “the resolve target”, simply because its not something I can take credit for, and it has nothing to do with my preferences - even though I do support and endorse its use/uptake. The concept has been put together largely as a function of ongoing discussions with Mad_Economist, Oratory1990 (this is where we have the DFHRTF from) and a few other scientifically minded folks in our community. I would credit them on this over myself.
Reviewer targets are running rampant these days, especially in the IEM space, and while that can be useful to get an indication of how an individual reviewer likes things to sound, it does cause problems for the collective understanding of this stuff. We need to endorse a view of the data that’s not merely a judgment against the preferences or HRTF/HpTF effects of a single individual, and this new paradigm we’re putting forward is a step towards course correcting on that.
It’s also a step towards providing a more complete picture of how a headphone performs, and while I wouldn’t go as far as to say “all the previous graphs are wrong”, I’m quite happy to say that at the very least they’re incomplete. And I think anyone who has actually gone through even some of the research on this topic is likely to agree that it makes sense to show the preference boundaries.
I think a more technically accurate term for this target (really its a new visualization method more than anything) would be the Harman RESEARCH target, since that’s what its based on. The DFHRTF starting point can be considered comparable to the pre-filter Harman in-room baseline with the same level of smoothing, and this means the 5128 data can be reasonably compatible with the preference research. The boundaries indicate the often missed preference groups (the segmentation paper for example) that have been established.
There is still some more work to be done here, since the 10dB tilt is a fairly coarse-grained view of what most people prefer.
Yup, that’s precisely what it is. A key goal of this project is to sort of… future proof headphone measurements - both for compatibility and for subsequent preference research that gets done. Say GRAS comes out with an even better HATS on the new standard set by the 5128 (I would not at all be surprised if they’re working on exactly this), we can apply the same methodology and have compatible results.
Right now, the biggest and most obvious advantage of the 5128 has to do with IEMs. I’ll say it bluntly… in my view, IEM measurements on 711 are just flat out missing key information. The acoustic Z difference is meaningful, and with the 5128 and 4620 we’re getting far better data for low and high frequencies as a result.
EDIT: I should note, the one change to the calibration we’re eventually going to work towards has to do with IEMs. Since IEMs bypass the pinna, it makes sense to use a population average for the pinna, rather than the one that’s part of the HATS. This isn’t something that exists yet, but I expect it’ll result in slightly less energy above 4khz for the IEM DFHRTF compensation - at least in comparison to the 5128 pinna.
IEMs are tough but the 5128 has been producing the best results so far, at least results that best reflect what I hear in the lows and the highs. Would be curious what you come up with.
Keep up the good work. You are far more erudite than me (I’m just cranky). I cannot state how happy I am that you’ve taken up this endeavor. Looking forward to what you’ve got coming up and willing to follow your lead.