Read further into that thread... it's not that simple.
Again, there's a lot more to it than that, and I'd encourage you to read through the rest of the discussion there. What he's talking about in the second bit is to do with their recent presentation that involved mannequin heads vs actual humans with canal blocked in-ear mics, and what they found is that mannequin heads in general aren't necessarily more accurate representations of humans, and this is in part due to them being more difficult to achieve a seal than on humans (and also some overestimations above 1khz as well). But that doesn't mean the 5128 isn't more 'humanlike' or somehow inferior to GRAS systems, merely that mannequin heads != closer results to measurements on humans. Now, if you look at the presentation... some of those mannequin 'prints' are... questionable, and likely don't have the same pinna deformation of official measurement rigs, and unfortunately it's not stated which result corresponds to which mannequin. So we'll need to do similar testing with in-ear mics on both humans and the 5128, and this is in part why our target is still a work in progress.
I already did.I'd encourage you to read through the rest of the discussion there.
I know.What he's talking about in the second bit is to do with their recent presentation that involved mannequin heads vs actual humans with canal blocked in-ear mics
I don't think anyone's saying it's inferior, just that there's no solid evidence it's superior, producing a more accurate approximation of the response on real heads than GRAS simulators.But that doesn't mean the 5128 isn't more 'humanlike' or somehow inferior to GRAS systems
I did. The presentation also shows none of the mannequin heads (of which the 5128 was one) did any better than flat plate GRAS rigs (45CA) with anthropomorphic pinnae (official GRAS KB5000 and Welti's custom ones which the Harman target is based on). It's all very well the 5128's ear canal being more anthropomorphic with a more accurate acoustic impedance, and looking 'humanlike', but if the head shape and very importantly hardness are all wrong resulting in excessive leakage and unrealistic reflections this is moot. That B&K (and GRAS) have completely failed to address the latter is baffling. Looks like you might need that skin-like ham for all your 5128 measurements after allmerely that mannequin heads != closer results to measurements on humans. Now, if you look at the presentation...
Testing on 2 (or even 3) subjects won't be enough. Sean's data above was from 15 subjects. The fact is you're not going to get as statistically robust results as Harman's multiple double-blind listening preference tests over the years that led them to their target.So we'll need to do similar testing with in-ear mics on both humans and the 5128
Summing up: mixing difuse field with sloping response that comes from a flat speaker in a room. Pretty much what I have described below.I’m genuinely curious.
Does anyone knows about any research done to relate the different measuring rigs?
Which is why we include the outcomes of the research for the target. And yeah, we're intending to do more in-ear mic measurements in general, even for evaluations from different people, but we can also find out fairly quickly if we need to cover the 5128 with ham haha.The fact is you're not going to get as statistically robust results as Harman's multiple double-blind listening preference tests over the years that led them to their target.
And as a consequence, imo, no way a flat measuring bass IEM like the SR Etymotics are representative of sloped response of speakers in a room, which should be the gold standard.
View attachment 254295
You also need to consider the SLD effect: acoustic sources at different distances from the ear are perceived to sound differently even if the measured response is the same. See Theile's seminal paper. Then there's the relative lack of tactile bass from headphones (and none at all from IEMs, which could, at least partially, explain the higher preferred bass for them) compared to speakers which needs to be compensated for. These are both reasons why preference listening tests are needed to arrive at the target, just as Harman have done.mixing difuse field with sloping response that comes from a flat speaker in a room