Was anybody at the Sean Olive talk at CanJam?

[deadkrillin]

What does "using" mean? As long as they throw out target curves that are not shown to be correct in human studies, they are producing worse results, not better. You can have the world's most accurate measurement rig but if you don't have a reference of what the response should be, it is for not.

I will say, the SoundGuys target was rated pretty positively in the new tests from Harman (arguably higher than Harman 2019 IE if you opt to weight trained listeners more), as well as in the SenseLab testing. Both studies have methodological issues, but I'm not quite as confident saying the SoundGuys curve in particular isn't assisting in producing useful data.

I think it's important to note that most of the targets based on some form of Harman—Harman IE 2019, Harman Beta 2024, HP.com's "target"—perform pretty well in this study. Arguably the biggest thing this study shows thanks to the Type 4620 is that the hallmarks of "good tuning" arrived upon in Harman's research are not only correct, but also (somewhat) agnostic to measurement fixture choice, HRTF/baseline choice, and transducer type.

In other words: this study goes some way to confirming that regardless of what you're using, Harman or something resembling/incorporating it is likely the best place to start.

 

[MayaTlab]

Funny, I suggested something similar a year ago: https://www.audiosciencereview.com/...target-you-can-try-the-eqs.43209/post-1624317

I'm not surprised either. Although I don't care about reviwer incentives as much, and I still think it might be not be preferred for a sizable percentage of listeners. Given that the individual FR variation of the IEM tested is as much as 10dB at 5kHz. I hope there is solution to handle conditions like these that's still fully data driven. At worst compliance in the top octaves will have to be done with some leniency when in situ FR is can have such high variability.
View attachment 356680

I largely agree, yeah. The data you've posted, as well as the fact that the majority of difference between the curves tested lies in the 4-8kHz area that a lot of people don't like about Harman's IE target, makes me think the MOA testing might do well to incorporate an adjustment band in this area. Can deal with the preference variance between listeners as well as anatomical variance all at once!

I'd encourage a different interpretation of that slide.

For a start it shows inter-individual variation at the eardrum depending on ear canal shape (length, volume, geometry) - based on a database of 20 ear canals from 10 individuals. However the eardrum impedance was kept a constant, and after looking at that slide and another recent article I'm definitely curious to see how varying eardrum impedance would affect the results.

The other thing is that if you consider that using someone's individual anatomy in a reference sound field (we'll go with someone's DF HRTF for now) is a desirable baseline from which to add preferential adjustments, then you would expect the DF HRTF at the eardrum to be different from individuals to individuals (and not just because of the pinna). Therefore, what you should be looking at isn't just the inter-individual variation at the eardrum, but rather the mismatch between how the DF HRTF varies at the eardrum (and if one wants to make comparisons with IEMs I think that it's a good idea to insist on the latter) of a set of individuals, and how an IEM's frequency response varies across that same set.

So now I'd love to know if Harman plans to measure (or has already measured !) the HRTF of the 3D printed busts of the individuals from the IHA database, with their ear canals terminated by the type 4.3 ear simulator's mic/eardrum, and compare it to the in-situ response of the IEMs, as well as test for varying eardrum impedance.

 

[markanini]

I'd encourage a different interpretation of that slide.

For a start it shows inter-individual variation at the eardrum depending on ear canal shape (length, volume, geometry) - based on a database of 20 ear canals from 10 individuals. However the eardrum impedance was kept a constant, and after looking at that slide and another recent article I'm definitely curious to see how varying eardrum impedance would affect the results.

The other thing is that if you consider that using someone's individual anatomy in a reference sound field (we'll go with someone's DF HRTF for now) is a desirable baseline from which to add preferential adjustments, then you would expect the DF HRTF at the eardrum to be different from individuals to individuals (and not just because of the pinna). Therefore, what you should be looking at isn't just the inter-individual variation at the eardrum, but rather the mismatch between how the DF HRTF varies at the eardrum (and if one wants to make comparisons with IEMs I think that it's a good idea to insist on the latter) of a set of individuals, and how an IEM's frequency response varies across that same set.

So now I'd love to know if Harman plans to measure (or has already measured !) the HRTF of the 3D printed busts of the individuals from the IHA database, with their ear canals terminated by the type 4.3 ear simulator's mic/eardrum, and compare it to the in-situ response of the IEMs, as well as test for varying eardrum impedance.

Interesting points but in a sense it doubles down on a single average curve being problematic. Maybe with larger data sets, afforded by modeling, the results can be weighted intelligently instead of "dumb" averaging, because there is existing data that begs the question of minimizing cases of understated 3-5kHz. We have equal loudness contours that show that those are ranges where the hearing is the most sensitive, therefore cases where excess energy happens will be worse than treating all deviations equally.

 

[IAtaman]

they used an off-ear headphone in the test which still interacts with the pinna so to get individual flat you need an in-ear mic. that is why I recommended IEMs

the way I understand it is that harman adds the ear-gain because listening to stereo speakers our ears expect that. but with binaural the ear-gain is baked into the recording itself. Flat FR for binaural is also what oratory1990 suggests & in my (biased) experience for binaural I preferred flat FR+IEMs

also take look at binaural recording devices (this one is neumann KU 100) and how mics are placed inside the ear-canal

That's is the point I think - what you said might be right, but the research you pointed at has nothing to do with that.

 

[IAtaman]

The talk has been posted on YouTube.

I think every public speech of Sean Olive should be recorded and made available to everyone interested.

Effectively a minor deity in audio world, yet as humble as ever.

I have seen first half of the slides before but I have never seen Dr. Olive present them. It was great to get a bit of the story of from him directly. When you are trying to learn calculus, it helps a lot to go back to the beginning and think about the problems Newton and Leibniz were trying to solve when they were developing this new tool. Watching Dr. Olive talk about his research was insightful in the same way.

The part about high frequency energy going down as more channels were measured was very surprising to me. Why does that happen?

 

[markanini]

I think every public speech of Sean Olive should be recorded and made available to everyone interested.

Effectively a minor deity in audio world, yet as humble as ever.

I have seen first half of the slides before but I have never seen Dr. Olive present them. It was great to get a bit of the story of from him directly. When you are trying to learn calculus, it helps a lot to go back to the beginning and think about the problems Newton and Leibniz were trying to solve when they were developing this new tool. Watching Dr. Olive talk about his research was insightful in the same way.

The part about high frequency energy going down as more channels were measured was very surprising to me. Why does that happen?

View attachment 356812

Probably cancellation, which affects HF more, due to smaller and more directional waves.

 

[staticV3]

The part about high frequency energy going down as more channels were measured was very surprising to me. Why does that happen?

Because speakers not aligned well with the 5128's pinnae rely more on sound reflected by the room to reach the eardrum.

Reflected sound is tilted warm, so as you include more and more surround speakers, the AVG also tilts warm.

The Heights are a worst case scenario for alignment, and therefore lack treble the most.

 

[MayaTlab]

The part about high frequency energy going down as more channels were measured was very surprising to me. Why does that happen?

View attachment 356812

What's a bit odd is the difference between blocked and open :

 

[IAtaman]

Because speakers not aligned well with the 5128's pinnae rely more on sound reflected by the room to reach the eardrum.

Reflected sound is tilted warm, so as you include more and more surround speakers, the AVG also tilts warm.

The Heights are a worst case scenario for alignment, and therefore lack treble the most.

Probably cancellation, which affects HF more, due to smaller and more directional waves.

Right, of course.

Initially, I thought adding more speakers should surely add more SPL even if most of the HF does not make it to the mics. But I understand these are normalized / level matched, so when level matched to L+R, 19 speakers have a warmer tilt because proportionally less HF reaches to the mics. Thanks.

 

[deadkrillin]

I'd encourage a different interpretation of that slide.

For a start it shows inter-individual variation at the eardrum depending on ear canal shape (length, volume, geometry) - based on a database of 20 ear canals from 10 individuals. However the eardrum impedance was kept a constant, and after looking at that slide and another recent article I'm definitely curious to see how varying eardrum impedance would affect the results.

The other thing is that if you consider that using someone's individual anatomy in a reference sound field (we'll go with someone's DF HRTF for now) is a desirable baseline from which to add preferential adjustments, then you would expect the DF HRTF at the eardrum to be different from individuals to individuals (and not just because of the pinna). Therefore, what you should be looking at isn't just the inter-individual variation at the eardrum, but rather the mismatch between how the DF HRTF varies at the eardrum (and if one wants to make comparisons with IEMs I think that it's a good idea to insist on the latter) of a set of individuals, and how an IEM's frequency response varies across that same set.

So now I'd love to know if Harman plans to measure (or has already measured !) the HRTF of the 3D printed busts of the individuals from the IHA database, with their ear canals terminated by the type 4.3 ear simulator's mic/eardrum, and compare it to the in-situ response of the IEMs, as well as test for varying eardrum impedance.

You are correct! I have absolutely no data on how eardrum impedance skews these results nor how that skew between listeners’ eardrums relates to the spread of variation between listeners’ HRTFs. Definitely would be an interesting thing to study.

 

[digitalfrost]

Harman 2024 Livestream with Dr. Olive, Crinacle, and Friends

 

[MacClintock]

A more sophisticated and accurate simulator is fine and great, but without the implementation of a proper scientific protocol to develop a reference target curve with it, I don't see the point of advertising alternatives as "superior" out of thin air. You first develop the instrument, do the research with it, and then you discover how the transducer is more or less supposed to be tuned. Because honestly it feels a bit "cargo culty" to present an alternative as superior, do the advertisement second, develop the instrument third and then do a research if any.

That doesn't mean that "artisan" alternatives are bad and can't be superior according to whatever relevant metric you choose. But good advertisement, marketing, influencing on social media, infotainment and "charisma" are not a replacement for proper research to validate any claim.

Sure, but apparently all the preparatory steps you descrived have now been done (at least by Sean Olive, not by crinacle or headphones.com). Or am I missing something?

 

[Dreamic]

Sure, but apparently all the preparatory steps you descrived have now been done (at least by Sean Olive, not by crinacle or headphones.com). Or am I missing something?

This was surely aimed at Headphones.com and company. Sean might have charisma (or maybe not if we look at the science nerds demographic) but "advertisement, marketing, influencing on social media, infotainment" he is too busy having a legitimate dignified profession as far as I know. Conflicts of interest are something an actual professional to be taken seriously understands. Headphones.com employees seemed perplexed by the concept when I tried to civilly discuss it with them out of curiosity in Discord.

 

[MacClintock]

Are the slides from Sean Olive's talk somewhere available? And the data from the new target?

 

[digitalfrost]

Are the slide from Sean Olive's talk somewhere available? And the data from the new target?

I traced the lines from the graph.

 

[Music1969]

Just to be clear. The BK5128 DF curve unmodified was the least preferred headphone target curve tested. This confirms previous listening tests where the DF wasincluded. It is too bright and thin, yet it remains the current standard and recommended target quoted in AES and IEC standards for the past 30 years.

The in-room measurements of the 19 loudspeakers is close to the DF target because the large number of sources coming from all directions approximate a DF and the speakers are equalized from anechoic flat to in-room flat meaning the speakers will also sound thin and bright.

If you apply a bass shelving and treble filter to the headphone matched to the flat in-room loudspeakers you can approximate the predicted-in room response of an anechoically flat loudspeaker. The downward slope of that curve is somewhere between 7-10 dB. There is a tolerance window around that target that will satisfy most people's taste. Some of the variance in taste is related to age (hearing loss), listening experience and possibly gender.

The HARMAN 2019 IE Target has a bass shelf boost and a treble shelf cut so is not DF tuned. The new HARMAN target we included in the test was the BK5128 DF with a bass and treble adjustment used in the HARMAN 2018 AE target: 6.6 dB bass shelf and -1.4 dB treble cut -- so it was not DF. I called it HARMAN Beta 2024 because it has not gone through any method of adjustment experiments where listeners modify the bass/treble to according to taste.

This post is the money shot.

What is AE Target?

I know Over Ear and In Ear but not A Ear

 

[staticV3]

What is AE Target?

Around Ear.

On GRAS, Harman has the IE2019 target, and the AE/OE2018 target.

It's just what they decided to call it. There's no deeper meaning.

 

[MacClintock]

View attachment 359054

I traced the lines from the graph.

Thanks a lot !

 

[markanini]

One big thing I take from this is that a target similar to Harman IE 2019, except approximately 5dB less energy between 3-8kHz, is statistically tied. In other words both targets are equally better than the others, despite significant differences in the upper midrange and lower treble. This should change how objective performance is evaluated moving forward.

 

[CedarX]

One big thing I take from this is that a target similar to Harman IE 2019, except approximately 5dB less energy between 3-8kHz, is statistically tied. In other words both targets are equally better than the others, despite significant differences in the upper midrange and lower treble. This should change how objective performance is evaluated moving forward.

That target (SoundGuys) also has approximately 3 dB less energy in the 20-70 Hz area per the above graph. Which of the two areas with the most difference matters the most for preference rating?

I think we could equally conclude that (paraphrasing): “One big thing I take from this is that a target similar to Harman IE 2019, except slightly less V-shaped and slightly more tilted, is statistically tied.”