Headphone EQ with in-ear microphones

[pkane]

I know there are a number of threads on this already, and maybe even one or two that I started, so I apologize in advance for yet another

I've done this type of measurement previously, and usually EQed it to a flat or slightly sloping target with good results. This time, I wanted to check to see how closely some of the artificial measurement rig error curves match my own, in-ear microphone measurements with HD650 (solid line is my FR error):

The things I note is less of an error in the lower frequencies than the Oratory or Inner-Fidelity measurements, and lack of the ~3KHz bump. What accounts for these differences other than possibly my own ear structure? Or is it the averaging over multiple positions?

My FR was captured with 4 averaged sweeps, each with me repositioning the headphone slightly to achieve as good a fit as I can make it by feel, as well as re-seating the microphone in the ear canal. The Harman target and the two error curves for HD650 were sourced from https://github.com/jaakkopasanen/AutoEq

 

[wasnotwasnotwas]

My brain forever gets confused here. Have I got this right as far as your 2 main areas of note:

1. Your measurements show your ears/ HP interaction dont roll off the lowest end as much as O/IF rig (maybe)
2. Likewise, your ears / HP interaction dont empashise the 3k as much as the O/IF rigs?

Or is it exactly the opposite ? So you would raise the bass and 3k more than their resulting EQ (to Harman or wherever)?

 

[pkane]

My brain forever gets confused here. Have I got this right as far as your 2 main areas of note:

1. Your measurements show your ears/ HP interaction dont roll off the lowest end as much as O/IF rig (maybe)
2. Likewise, your ears / HP interaction dont empashise the 3k as much as the O/IF rigs?

Or is it exactly the opposite ? So you would raise the bass and 3k more than their resulting EQ (to Harman or wherever)?

The error should be a flat line at 0. Where it's further away from zero in my measurement, it requires more correction to bring it back to zero. Where it's closer to zero, it requires less correction.

The 3k bump is more pronounced in the artificial rigs than in my testing, but the reason for that is likely lack of resonance due to the blocked ear-canal in my test. So, in reality, HD650 may need a larger correction at 3k towards zero than my measurements might indicate.

 

[Dreyfus]

The error is due to the lack of ear canal resonance. Here you can see the specific components of the ear-realated response specified by G.R.A.S.:

Do keep in mind that most in-ear mics are also skipping the concha gain unless you are using really small capsules and stick them right into the ear canal. I use the Primo EM258 from time to time. Doing measurements on the "original head" can be a tough quest though with both the headphone and capsule moving around. Dummy head recordings offer much better repeatability after all. However, I have had pretty good success equalizing bass and mids to flat with in-ear measurements. Since the highs should only be evaluated with the ear canal resonances involved, I tune the upper frequencies by ear only, listening to white or pink noise. Much better than relying on a "generic" dummy head approximation, imo.

Regards
Dreyfus

 

[pkane]

The error is due to the lack of ear canal resonance. Here you can see the specific components of the ear-realated response specified by G.R.A.S.:

View attachment 94114

Do keep in mind that most in-ear mics are also skipping the concha gain unless you are using really small capsules and stick them right into the ear canal. I use the Primo EM258 from time to time. Doing measurements on the "original head" can be a tough quest though with both the headphone and capsule moving around. Dummy head recordings offer much better repeatability after all. However, I have had pretty good success equalizing bass and mids to flat with in-ear measurements. Since the highs should only be evaluated with the ear canal resonances involved, I tune the upper frequencies by ear only, listening to white or pink noise. Much better than relying on a "generic" dummy head approximation, imo.

Regards
Dreyfus

Pretty much what I thought. I do find it interesting that we often try to equalize using data from a simulation of some sort of "average" human ear, while the individual ear structure can be very different. My desire was to try to get to a measurement of my own ear FR to see if this can be better compensated. So far, my subjective preference is to equalize the FR of the headphones with in-ear mics to a simple downward-sloping line of -5dB from 20Hz to 20KHz. This sounds the best to me. Compensating to Harman's curve, instead, produces a result that's too bright and lacks bass. I tried to adjust this by bumping up bass response, and it helps a little, but the overall sound isn't as balanced or natural as when I EQ to a simple sloping line. FWIW.

 

[Dreyfus]

Well, measuring with an averaged ear is pretty much the only reasonable thing we can do when we want to present data to a large audience AND tune headphones to deliver a pleasing tonality to the mass market. After all, even the Harman curve is just a broad approximation since it has been derived from both measurements with averaged ears (G.R.A.S. pinna and coupler) and averaged preference ratings of listening groups. The best we should expect here is a decent tonal balance that is at least not totally of from our natural hearing.

On the other hand, fixing HRTF issues and matching the performance to the individual ear is a totally different story, indeed. The things we have to address are not only the standard deviations on the broad spectrum, like different tastes for bass boost or varying concha/pinna gains. Ideally we would also have to make sure that the narrow peaks and troughs above 5 kHz are at the right place and level so they match the spectral maps we experience in real life (see the "gestalt" model by Theile for further explinations).
Fixing all of that is actually not that easy because we have no standardized and - much more important - practically convenient procedure to evaluate the actual frequency response at the ear drum of the individual. Measuring with in-ear microphones at the blocked ear canal is flawed because we skip the individual canal resonances. Also, it is hard to find a valid microphone calibration that addresses the pressure build up in front of the capsule which may lead to up to 6 dB of treble boost depending on the propagation (concentration) of the sound field. Such effects may vary with driver size, angle and distance, which - as a side effect - may also excite deviating pinna and ear canal resonances.

To summarize, it is not only hard to come up with a convenient solution that works well for the individual. It is also hard to ensure good repeatability with varying headphone designs as much as the varying sealing and seating incidences of both headphone and microphone on/in the ear.

All of those difficulties form the reason why I do not compensate above 2 kHz with in-ear measurements. As for the treble response I only trust in manual adjustments done with white/pink noise by ear. To my experience, mids and lows are pretty safe to equalize with in-ear mics, though. I tend to like a flat response from 20 Hz to 2 kHz. Maybe with a tad of bass boost on top. Harman is too much for me, though. IMO there is no replacement for body (bone) resonance. Simply pushing up the bass does only bloat things up and increase the distortion.

 

[pkane]

Well, measuring with an averaged ear is pretty much the only reasonable thing we can do when we want to present data to a large audience AND tune headphones to deliver a pleasing tonality to the mass market. After all, even the Harman curve is just a broad approximation since it has been derived from both measurements with averaged ears (G.R.A.S. pinna and coupler) and averaged preference ratings of listening groups. The best we should expect here is a decent tonal balance that is at least not totally of from our natural hearing.

On the other hand, fixing HRTF issues and matching the performance to the individual ear is a totally different story, indeed. The things we have to address are not only the standard deviations on the broad spectrum, like different tastes for bass boost or varying concha/pinna gains. Ideally we would also have to make sure that the narrow peaks and troughs above 5 kHz are at the right place and level so they match the spectral maps we experience in real life (see the "gestalt" model by Theile for further explinations).
Fixing all of that is actually not that easy because we have no standardized and - much more important - practically convenient procedure to evaluate the actual frequency response at the ear drum of the individual. Measuring with in-ear microphones at the blocked ear canal is flawed because we skip the individual canal resonances. Also, it is hard to find a valid microphone calibration that addresses the pressure build up in front of the capsule which may lead to up to 6 dB of treble boost depending on the propagation of the sound field (concentration vs diffusion). Such effects may vary with driver size, angle and distance, which may also excite deviating pinna and ear canal resonances.

To summarize, it is not only hard to come up with a convenient solution that works well for the individual. It is also hard to ensure good repeatability with varying headphone designs as much as the varying sealing and seating incidences of both headphone and microphone on/in the ear.

All of those difficulties form the reason why I do not compensate above 2 kHz with in-ear measurements. As for the treble response I only trust in manual adjustments done with white/pink noise by ear. To my experience, mids and lows are pretty safe to equalize with in-ear mics, though. I tend to like a flat response from 20 Hz to 2 kHz. Maybe with a tad of bass boost on top. Harman is too much for me, though. IMO there is no replacement for body (bone) resonance. Simply pushing up the bass does only bloat things up.

Thank you for the thorough answer. I do agree with what you're saying. The question I don't know the answer to is whether the blocked ear canal with in-ear mics introduces a larger error than the difference between artificial ear/head constructs and my actual ears. In the comparison I posted, it would seem that the differences are not huge, with the few deviations being the lack of measured resonance around 3kHz and a slightly greater bass response below 200Hz. Perhaps these can be accounted for by some standard compensation. What I'm trying to get to is a more "personalized" headphone EQ, if that's even possible. From my personal preference stand point, I do like the flat line, down-sloping EQ with in-ear mics with both, HD650 and HE560 headphones (also old Stax Lambda). But that may just have to do with my preferences and different ear structure, so I can't generalize.

 

[Dreyfus]

Who knows. The only way to find out is probably to measure, equalize and compare.

IMO there is nothing wrong with applying a warm tilt to a linearized response. That is similar to some speaker house curves and may equally be valid for headphones as long as you don't overdrive the effect and don't rely on the accuracy of the treble (for the reasons explained above).

I wouldn't expect much from a generalized compensation curve to match your findings to the fancy G.R.A.S., gear by the way. Have a look at Jaakko Pasanen's document about the errors between the well-respected systems used by Oratory, Crinacle, Headphone.com, InnerFidelity, Rtings etc.:
Comparing Headphone Measurement Systems
I do not fully agree with his notion that a single source should be lifted out as a gold standard to allow a judgement about other equipment (but ok, you have to start somewhere...). Still, you can see that even standardized systems show some noteworthy errors between varying sources. The crux is that every headphone interacts differently with every specific pinna. That is why the industry tries to found a standard (average) pinna that induces less deviations across the board - instead of relying on highly individual pinnas. And then there is placement, seal and unit-to-unit variation. All of those aspects can render a generalized compensation curve completely useless. For those reasons, don't expect too much.

Regards
Dreyfus

 

[bobbooo]

Compensating to Harman's curve, instead, produces a result that's too bright and lacks bass.

What EQ settings are you using to compensate to the Harman target?

 

[pkane]

What EQ settings are you using to compensate to the Harman target?

You can see the difference curve in my original post: that’s the difference between Harman curve and my measured FR. That’s what I’m correcting to a flat 0 line.

 

[bobbooo]

You can see the difference curve in my original post: that’s the difference between Harman curve and my measured FR. That’s what I’m correcting to a flat 0 line.

Have you tried just using Oratory's EQ settings for the HD650? (AutoEQ's settings do not correctly target the Harman curve by the way, but intentionally around 2-3dB too low in the bass, plus other differences). Oratory uses the exact same measurement rig as Harman did when creating their target, so will produce EQ settings most accurate to Harman. If you have tried his settings, did you still find it to sound too bright and lacking bass?

 

[pkane]

Have you tried just using Oratory's EQ settings for the HD650? (AutoEQ's settings do not correctly target the Harman curve by the way, but intentionally around 2-3dB too low in the bass, plus other differences). Oratory uses the exact same measurement rig as Harman did when creating their target, so will produce EQ settings most accurate to Harman. If you have tried his settings, did you still find it to sound too bright and lacking bass?

Oratory's EQ settings sound even brighter than my own EQ corrections towards Harman. Bass is better, though.

 

[andreasmaaan]

Oratory's EQ settings sound even brighter than my own EQ corrections towards Harman. Bass is better, though.

So interesting how different people's perceptions of the Harman headphone curve can be. I use the Oratory settings for HD650 (well, mine are HD6xx technically), and I find it perfectly matches my perception of neutral speaker in-room sound, apart from having way too much bass. So I add a -5dB bass shelf below about 120Hz, which brings the balance to somewhere between Harman target and diffuse field, I guess.

 

[pkane]

So interesting how different people's perceptions of the Harman headphone curve can be. I use the Oratory settings for HD650 (well, mine are HD6xx technically), and I find it perfectly matches my perception of neutral speaker in-room sound, apart from having way too much bass. So I add a -5dB bass shelf below about 120Hz, which brings the balance to somewhere between Harman target and diffuse field, I guess.

Certainly there are variations in hearing, ear structure and resonances, and possibly even between two sets of the same type of headphones. Which is why I've been trying to figure out if I can EQ things to my own liking, accounting for most of these variables

 

[andreasmaaan]

Certainly there are variations in hearing, ear structure and resonances, and possibly even between two sets of the same type of headphones. Which is why I've been trying to figure out if I can EQ things to my own liking, accounting for most of these variables

Absolutely

I also wonder whether relative levels of bass are a largely personal preference when it comes to headphones. There was certainly much greater deviation in preferred bass level than preferred treble level in the Olive headphone preference study. And I can also imagine that some of us have different intuitions about how much bass a headphone should reproduce (ranging possibly all the way from diffuse field, to Harman target, and maybe even right through to wanting extra bass to compensate for the lack of tactile sensation).

 

[thewas]

Also we shouldn't forget that chosen bass level also depends a lot on the listening level and unfortunately also on the recordings used.

 

[bobbooo]

Plus a whole host of subconscious cognitive biases associated with sighted listening, and even those that may remain in conventional blind listening tests, such as a 'more/less is better' bias.

 

[bobbooo]

maybe even right through to wanting extra bass to compensate for the lack of tactile sensation

I suspect this plays a significant role in headphone bass preference.

 

[pkane]

Plus a whole host of subconscious cognitive biases associated with sighted listening, and even those that may remain in conventional blind listening tests, such as a 'more/less is better' bias.

There's no question that the curves are different enough that they can be distinguished in a blind test (and this was confirmed). In some very audible parts of the spectrum the differences are more than a couple of dB. In the case of comparing Oratory and a straight-line EQ, straight line is mostly "less". So maybe I prefer less than more?

In any case, this was a preference test, i.e., subjective, to determine my preferences only, and it was done blind.

 

[bobbooo]

There's no question that the curves are different enough that they can be distinguished in a blind test (and this was confirmed). In some very audible parts of the spectrum the differences are more than a couple of dB. In the case of comparing Oratory and a straight-line EQ, straight line is mostly "less". So maybe I prefer less than more?

In any case, this was a preference test, i.e., subjective, to determine my preferences only, and it was done blind.

It's not a discernable difference test, it's a preference test as you say. But if you knew what the two EQ curves were that you were comparing beforehand, it wasn't a fully blind test. As soon as you recognise which is which, the test effectively stops being blind and all subconscious biases about each curve are on the table again. For it to be fully blind someone else would have had to create and switch between the EQ curves without you seeing what they were beforehand, and better still not even tell you what kind of change there is between the tests, so you wouldn't even know you're listening to EQ differences.