- Joined
- Jun 4, 2014
- Posts
- 3,652
- Likes
- 2,927
It seems it's open season on the Harman target. Every reviewer and his dog now has something 'better'. And then there's this epic rant:
Despite the fact I wish hawaiibadboy wouldn't say '****' quite as much (I guess he has to keep up the bad boy persona?), he's actually one of the few reviewers worth watching because he doesn't just endlessly shill a supply of free review samples. But I need to push back a bit on his main claim that 249 people isn't a large-enough sample size. We simply can't know what the confidence intervals would be without at least knowing the variance from that sample. One thing's for sure, it's a sample size 248 data points larger than any other reviewer's 'improvement'. As individuals, our preference targets might not perfectly correlate with Harman, but that's not the point of the Harman target. Yes, we'd want a better curve for each of us individually (more on this below), but it's also interesting and relevant to know what the average preference would be. I'm perfectly happy that outliers like the RHA CL2 exist - life would be boring if everybody was doing a Moondrop and trying to perfectly match the same target with every single headphone ever released - however, averages should be relevant to us, and to OEMs, especially if they're trying to appeal to the widest-possible market.
Here's one curious observation (some of you may even be able to try this at home). Take any recent IEM measurement database (anybody's), and average the entire set and see what graph shape you come up with. It won't be that different from the Harman target - and I don't think this is a coincidence. Even if a headphone manufacturer didn't know what the Harman target was, it's a safe bet they would at least try to build/tune something that sounded reasonable - at least to them. If you average enough samples, outliers get effectively filtered out.
So I have a request for any interested headfiers. It's not going to be easy get 250+ people together in an anechoic chamber with perfect SPL-control and an arbitrary parametric eq system during a pandemic, but I believe there's a way of approximately establishing individual and (with enough data points) average preference curves. This is potentially useful not only for verifying/improving existing target functions, but it would also allow us to establish equivalent targets for other ear simulators, such as the newer B&K 5128 couplers.
The idea is as follows. Simply provide a short list of your absolute, money-no-object, all-time favorite IEMs (I'd recommend 2<=x<=5). No need for the usual debate about who's right and who's wrong on the choice of 'best-sounding' IEM; we should expect variance. We just want to see those headphones listed. If you can find those headphones on this database: https://www.hypethesonics.com/iemdbc/ then you can take the next step which is to load them up (only those select-few favorite headphones) and then hit the average button in the toolbar (it appears only when x>=2):
If you don't find your absolute-favorite headphone(s) listed, just make a note of that and we'll see if we can get it added.
In anticipation of the push-back, hate mail, etc., I'll acknowledge this approach won't be perfect. Two headphones you really like might average together to create a hybrid monster you don't. But that's ok. We're only talking about averages for now. The more problematic issue is that limitations of existing headphones might give the impression that we all enjoy less sub-bass, more mid-bass, resonance peaks somewhere between 7 kHz - 11 kHz and a rapid roll-off beyond that. But many newer IEMs are making strides in terms of better extension at both ends of the spectrum and avoiding pronounced resonance peaks, so there's some hope we won't be that far off. The obvious next step for the individual user is that you can then take your own average and use that (instead of the Harman target) as an EQ target, or to search for other IEMs you might enjoy. The algorithms for this still won't be perfect, because even with the perfect target curve there's a question of what weight we apply to each part of the spectrum. But it's a start.
I'll get the ball rolling
Let's hypothetically say you have a listener that enjoys the Shure KSE1500, Beyerdynamic Xelento, special black edition of the Vision Ears Erlkönig and FiR Audio's M5. Then you'd end up with an average 'target' curve that looks like so:
Given the provisos above, it's still actually not that far off the Harman target. But again, this an insignificant sample size of one. So feel free to help us expand on that...
Despite the fact I wish hawaiibadboy wouldn't say '****' quite as much (I guess he has to keep up the bad boy persona?), he's actually one of the few reviewers worth watching because he doesn't just endlessly shill a supply of free review samples. But I need to push back a bit on his main claim that 249 people isn't a large-enough sample size. We simply can't know what the confidence intervals would be without at least knowing the variance from that sample. One thing's for sure, it's a sample size 248 data points larger than any other reviewer's 'improvement'. As individuals, our preference targets might not perfectly correlate with Harman, but that's not the point of the Harman target. Yes, we'd want a better curve for each of us individually (more on this below), but it's also interesting and relevant to know what the average preference would be. I'm perfectly happy that outliers like the RHA CL2 exist - life would be boring if everybody was doing a Moondrop and trying to perfectly match the same target with every single headphone ever released - however, averages should be relevant to us, and to OEMs, especially if they're trying to appeal to the widest-possible market.
Here's one curious observation (some of you may even be able to try this at home). Take any recent IEM measurement database (anybody's), and average the entire set and see what graph shape you come up with. It won't be that different from the Harman target - and I don't think this is a coincidence. Even if a headphone manufacturer didn't know what the Harman target was, it's a safe bet they would at least try to build/tune something that sounded reasonable - at least to them. If you average enough samples, outliers get effectively filtered out.
So I have a request for any interested headfiers. It's not going to be easy get 250+ people together in an anechoic chamber with perfect SPL-control and an arbitrary parametric eq system during a pandemic, but I believe there's a way of approximately establishing individual and (with enough data points) average preference curves. This is potentially useful not only for verifying/improving existing target functions, but it would also allow us to establish equivalent targets for other ear simulators, such as the newer B&K 5128 couplers.
The idea is as follows. Simply provide a short list of your absolute, money-no-object, all-time favorite IEMs (I'd recommend 2<=x<=5). No need for the usual debate about who's right and who's wrong on the choice of 'best-sounding' IEM; we should expect variance. We just want to see those headphones listed. If you can find those headphones on this database: https://www.hypethesonics.com/iemdbc/ then you can take the next step which is to load them up (only those select-few favorite headphones) and then hit the average button in the toolbar (it appears only when x>=2):
If you don't find your absolute-favorite headphone(s) listed, just make a note of that and we'll see if we can get it added.
In anticipation of the push-back, hate mail, etc., I'll acknowledge this approach won't be perfect. Two headphones you really like might average together to create a hybrid monster you don't. But that's ok. We're only talking about averages for now. The more problematic issue is that limitations of existing headphones might give the impression that we all enjoy less sub-bass, more mid-bass, resonance peaks somewhere between 7 kHz - 11 kHz and a rapid roll-off beyond that. But many newer IEMs are making strides in terms of better extension at both ends of the spectrum and avoiding pronounced resonance peaks, so there's some hope we won't be that far off. The obvious next step for the individual user is that you can then take your own average and use that (instead of the Harman target) as an EQ target, or to search for other IEMs you might enjoy. The algorithms for this still won't be perfect, because even with the perfect target curve there's a question of what weight we apply to each part of the spectrum. But it's a start.
I'll get the ball rolling
Let's hypothetically say you have a listener that enjoys the Shure KSE1500, Beyerdynamic Xelento, special black edition of the Vision Ears Erlkönig and FiR Audio's M5. Then you'd end up with an average 'target' curve that looks like so:
Given the provisos above, it's still actually not that far off the Harman target. But again, this an insignificant sample size of one. So feel free to help us expand on that...
Last edited:
