Measuring HRTF for headphone use

[pkane]

Welp, I tried your method @pkane and here is the resulting curve I got by measuring my headphones using the Dayton Audion imm-6:
View attachment 15448

I then took the inverse to make it flat for both left and right. It sounds ok, but isn't as good as I hoped. Maybe cause I'm treating this measurement as I would with my speaker measurement, which I assume is the wrong way to got about that since headphones and speakers perform differently?

Sorry, I’m confused. You measured your headphones using imm-6? How did you do that? And when you say you inverted the measured curve, how did you apply this to music?

 

[Shikamon]

All I did was measure each side of my headphones in air using the imm-6. I had a feeling it may not work cause it's not on my head... And then I took the curve measured and applied the inverse of 100hz to 20khz through Equalizer Apo. The sound results was not to my liking and I expected much since I bet I would need in ear mics like yours to do it. :0

 

[pkane]

All I did was measure each side of my headphones in air using the imm-6. I had a feeling it may not work cause it's not on my head... And then I took the curve measured and applied the inverse of 100hz to 20khz through Equalizer Apo. The sound results was not to my liking and I expected much since I bet I would need in ear mics like yours to do it. :0

Yes, measuring in the air isn’t going to be accurate. Did you re-measure after applying the inverted curve? Correctly done, inverting a curve will result in a flat response. I do suggest you try for a tilted one. Do you use parametric EQ? What headphones?

 

[Shikamon]

Yes, measuring in the air isn’t going to be accurate. Did you re-measure after applying the inverted curve? Correctly done, inverting a curve will result in a flat response. I do suggest you try for a tilted one. Do you use parametric EQ? What headphones?

Well I just re-measured my SHP 9500S after compensating and it's close to be flat:

I'll try adding a -10 slope but I doubt it will do much good compared to my previous method...

 

[Shikamon]

I just did the -10B slope and it still sounded really dark and less detailed....
Here is what my Equalizer APO applied curve looks like:

 

[pkane]

I just did the -10B slope and it still sounded really dark and less detailed....
Here is what my Equalizer APO applied curve looks like:
View attachment 15450

Strange. It appears that the curve doesn’t correct under 100hz but is very aggressive above and up to 20KHz. I usually focus on correcting from 25-30Hz to about 10Khz and leave the rest up to the ‘natural’ headphone curve. But I’d measure it with a sealed cup over the ear, as open air measurements will not produce the correct low frequency response.

 

[Shikamon]

Yeah. I'll never get that with the mic I have. So I guess Greisinger's method works best for me I guess.

 

[solderdude]

It's all a matter of references/calibrations and there are unknowns which are assumed to be known.

Griesingers method stands and falls on pitfalls as well.
One of them is using your ears as analyzers.
When you read the line on the bottom of my post you will get my P.O.V.

One would have to understand that Griesinger makes it look easy but there is one aspect that is important here. Maybe even 3 or 4.

One aspect is experience.. age and hearing damage are incorporated in the test.
The second aspect is the speaker used. It is a small one in close proximity and EQ'ed by the same experienced ears as well.
This must be a point source at short distance which 'normal speakers' aren't. Has to be a small and good one. I really doubt a $ 15.- speaker can reproduce 30Hz at the proper level, even at that distance. IMO a good speaker is needed. You see him 'do it' with an expensive one.
What's important here as well is the distance, as it is far away from walls. You will only hear the speaker and not so much the room.
The 3rd aspect here is SPL level. The EQ will certainly differ when doing the test as different levels. At least the bass levels will.
An EQ found for 70dB SPL will differ when done at 80dB SPL.
An SPL meter will easily be able to tell you, you need one when doing the mic thing anyway to calibrate the measurement chain.
The 4th aspect is that you must not use sinewaves but must be done with noise bands. Sinewaves will certainly give you the wrong corrections.

Then comes the last bit... The arrived EQ is personal and has no bearing with measurements of the headphone unless you measure it on a calibrated rig. Just assuming it is calibrated when buying a mic that measures 'flat' under unknown conditions doesn't mean anything.
I would simply forget the desire to compare findings obtained with Griesingers method with measurements.
The reason is SPL and 'preference' as well as 'experience' with listening to noise bands is essential.

Measurements are not SPL dependent and do not incorporate ear damage unless you calibrate on speakers in a room that 'measure' kind of flat at the listening position. You can do it with noise bands. Sennheiser does in the plots they publish of the HD800 etc but purposely limit the bandwidth of the measurements.
When using Griesingers method at a certain SPL the 'slope' is already incorporated but not with mics (assuming calibration is properly done)
When using mics (and sweeped sinewaves) I would recommend to smooth in REW using 'acoustic' smoothing

Using Griesingers method with normal speakers in a room at listening position (using calibrated by in-ear mics) and any preferred 'slope' added (most look alike but differ here and there a dB or so) and then measurements with the same mics using headphones might yield quite different results in obtained correction curve. These cannot be compared.

I do like the exercises done by both of you and in all cases will add to your personal insight in your hearing abilities and preferences.
There is something to be said for both methods (measuring with mics or by ear) but the results of both methods will not yield the same corrections.
One stands or falls with the correct execution (by ear) copied exactly in Griesingers method and experience with noisebands and knowing what equal loudness is where the other one stands and falls with mic-measurement chain calibration and applied 'preference curves'.

What I think pkane is doing here is a more 'combined' method (Griesinger and mics) which is an interesting approach.

Just my P.O.V.

 

[Shikamon]

I see @solderdude. But what are noise bands anyway? I thought using sine waves are enough to do the experiment?

Kinda getting more questions then answers at this point....

And doing @pkane way of compensating headphones is something I would like to try if I can find and purchase some good in ear mics. Any good source to find some? :0

 

[solderdude]

The difference between a sinewave and a noise band (what Griesinger is using) is that within a very narrow band (the one where you just might happen to set your test sinewave at) there can be substantial peak or dip (as in a few dB).
Look at some of my measurements, which are not smoothed for that reason, where you can see such dips.
Such local dips (and peaks) can skew your EQ substantially where when you would be using a noise band this risk is much smaller due to the random nature of all the frequencies in that band are averaged both by your ear and mic when doing it that way.

You can even make measurements using noise bands and get 'smoother' plots because of it. Sennheiser does this for instance.

When using an all microphone approach it is different as you can smooth those dips away and see them easily.
The question remains, however, how accurate that mic is under those circumstances as it does not seem to make much use of Concha gain and blocking of the Tragus as it sits a bit outside of the ear canal and part of the Concha is filled with plastic.
It might well make accurate measurements but you would have to apply corrections for the lower frequencies afterwards (or apply a room curve if that is more to your preference).
That's what I was curious about the HD650 measurements from pkane as the HD600 and HD650 have almost no Pinna 'activation' and should measure pretty much the same with or without a Pinna.
Would be very interested to see those results.

 

[Shikamon]

The difference between a sinewave and a noise band (what Griesinger is using) is that within a very narrow band (the one where you just might happen to set your test sinewave at) there can be substantial peak or dip (as in a few dB).
Look at some of my measurements, which are not smoothed for that reason, where you can see such dips.
Such local dips (and peaks) can skew your EQ substantially where when you would be using a noise band this risk is much smaller due to the random nature of all the frequencies in that band are averaged both by your ear and mic when doing it that way.

You can even make measurements using noise bands and get 'smoother' plots because of it. Sennheiser does this for instance.

When using an all microphone approach it is different as you can smooth those dips away and see them easily.
The question remains, however, how accurate that mic is under those circumstances as it does not seem to make much use of Concha gain and blocking of the Tragus as it sits a bit outside of the ear canal and part of the Concha is filled with plastic.
It might well make accurate measurements but you would have to apply corrections for the lower frequencies afterwards (or apply a room curve if that is more to your preference).
That's what I was curious about the HD650 measurements from pkane as the HD600 and HD650 have almost no Pinna 'activation' and should measure pretty much the same with or without a Pinna.
Would be very interested to see those results.

Awesome! I get it now! xD When I was listening to the video Griesinger did on Youtube, I did notice the tones he used were noise bands and not sine waves. I just thought of something too. What I did I opened up REW and played pink noise through my headphones. Then, I opened up Peace Gui (a UI addon for EQing with Equalizer APO) and divided each bar to be 1/3 octave apart and reduced the volume for everyone except 500hz (my reference band). Then, I went up each volume bar to get it sound similar in loudness to my reference 500hz until I get to 16khz and vice versa down to 10hz. The result sound pretty darn smooth compared to what I had previously using sine waves.

I'm going to try this with my speakers too and see where that gets me.

 

[solderdude]

When 'neutral' is done properly it should sound very good on well made recordings but on the downside will expose poorly recorded material ruthlessly. When something isn't really flat but only appears to measure flat (because of measurement gear not being as neutral as assumed it is) then it won't sound convincingly real (IMO).
So when it sounds really good to you chances are it is good (for that specific loudness) as explained in the 'Fletcher-Munson' part.
A little bass boost on top of 'flat' is needed (for me) when using headphones.
Just linking because it is too much text and illustrations to copy in here.

 

[Shikamon]

When 'neutral' is done properly it should sound very good on well made recordings but on the downside will expose poorly recorded material ruthlessly. When something isn't really flat but only appears to measure flat (because of measurement gear not being as neutral as assumed it is) then it won't sound convincingly real (IMO).
So when it sounds really good to you chances are it is good (for that specific loudness) as explained in the 'Fletcher-Munson' part.
A little bass boost on top of 'flat' is needed (for me) when using headphones.
Just linking because it is too much text and illustrations to copy in here.

Good explanation @solderdude! I just went through doing the equal loudness test using noise bands instead of sine waves and the sound is a lot better! I'm listening to a live commentary via online and they sound pretty realistic, especially with the bass boost I've added in. Thanks for pointing out the importance of noise bands. Our ears are meant to listen to music compared to listen to sine waves, which are mostly meant for measurements via microphone.

 

[pkane]

That's what I was curious about the HD650 measurements from pkane as the HD600 and HD650 have almost no Pinna 'activation' and should measure pretty much the same with or without a Pinna.
Would be very interested to see those results.

@solderdude, here is a measurement using an in-ear mic and HD650. It's actually much better behaved and much closer to the shape of -10dB sloped line than the measurements I was able to get using an external mic. Also, much better behaved than the Innerfidelity raw measurements. I wonder if that's why HD650s are so popular -- they don't need much correcting to get to the frequency response that Olive found is preferred by the majority of listeners.

 

[solderdude]

As a comparison here is my measurement of a recent HD650 (its compensated in the lows)
A horizontal line is 'audible flat' to at around 70-80dB SPL.
In your case the blue line is audible flat.

This is how they sound to me as well (at that volume). Somewhat lacking in the subbass apartment with a hint of 'warmth'(hump around 150Hz) and not needing any correction at all in the mids and treble.
That is indeed why it is universally loved and still today is one of the most neutral headphones.
Add to that there is virtually no Pinna activation which means that ear shape etc. has little influence on the perceived SQ
Downside (which does not show) is the 'narrow' headstage it has but does not bother me.

When I EQ mine (an older HD650) it measures like this:

Note: the frequency limits and sizing of the plots differ.
The realism is scary. There is a great rumble, not artificial warmth but just enough and the treble is sparkly, detailed and subdued with plenty of 'air' but not 'ethereal' sounding nor 'hyper-detailed', just the right amount.


Let's analyze some more and welcome everyone in the world of 'measuring headphone science' (the call for a standardization plea ?)

When looking at your plot I would conclude the HD650 is 'flat' down to 30Hz (for me it is 50Hz almost one octave higher)
Around 150Hz you have about +2dB which is the same as mine.
Both are on reference at 1kHz.
There is an obvious difference around 2.5kHz where I measure -3dB and you measure +4dB so a 7dB discrepancy.
This is obviously a Concha effect which you measure and I don't.
Knowing the Concha can make a difference I always 'listen' to headphones as well to determine whether or not a dip in that area is audible or not. That part of the frequency band is where 'presence/clarity' in sound is. Not to be confused with 'forward' sound which is at a lower band around 2kHz. The plot below shows what I think something sounds like when those bands are pushed up or down on an equalizer when coming from 'neutral = realistic' sound.

Let's take other measurers plots (everyone has measured a HD650) and let's use those with real HATS (with Pinnae) and that applied compensation already.

Below Sam Vafei's (Rtings) attempt

Interestingly his measurement shows no bass extension at all, it has some to me, just not the rumble on the correct level.
His measurements show a drop off from the 150Hz point and is 3 octaves higher than mine.. seems incorrect to me.
At 30Hz he is at -7dB, I am -3dB and yours is 0dB. But his measurements (with Pinna) also show +1 to +3dB.

Here are his 'without Pinna' measurements (which is referenced to the horizontal grey line) and 'with Pinna" measurements (the blue line) which show the Pinna effect (the grey dotted is his 'target' curve) and around 2.5kHz he measures +4dB on Pinna activation.

When I would add that to my measurement it would show +1dB at 2.5kHz. It is why I don't compensate at all in that area in my EQ.
When you would compensate to the blue line you would end up with a -1dB dip there (a small dip is preferred over a small peak)


Tyll (innerfidelity in its better years), unfortunately, applied the wrong compensation (above 1kHz) from day one and kept using it till he resigned as to keep the plots comparable.
His plots thus not reflect a horizontal line as 'flat' above 1kHz.
I drew in a 'reference' line (in green) and when you de-tilt that green line to a flat line it displays the perceived tonal balance better.
My green line is to be viewed as your blue line. The green line is not the exactly correct line as all of his plots show a peak at 10kHz which is incorrect but it is closer to the truth.

His plots show 40Hz (at 1kHz ref) which is nicely between ours.
He too has a +2dB hump at 150Hz and references to 1kHz. At 3kHz he also has a small bump (look at de-tilted plot below the real plot)
At 10kHz he has a peak of around +3dB where non of us have this it should be just below 1kHz level as all our plots show.

Above 10kHz things are not reliable anymore with a Pinna. Mine is measuring the driver response directly and think is 'more trustworthy' as there are no Pinna interactions.


Below the attempt of our Russian friends (RAA) using a HATS

It starts to drop off (with respect to 1kHz) at 50Hz and has a +2.5dB hump at 150Hz. At 2.5khz it shows -1dB.
The weird part (and is why I think their compensation is really 'off') is that at 5kHz it measures +6.5dB !
That would make it terribly sibilant which it is not at all.
A -10dB dip at 10kHz also seems awfully wrong.


Headroom uses Tyll's data but with better compensation (at least when set to 'frequency response compensated' and '20dB')
http://graphs.headphone.com/graphCompare.php?graphType=-1&graphID[]=853&scale=20
(direct links are not allowed ?)
Extends to 22Hz (ref 1kHz) which I don't believe it does, and shows a +5dB hump around 150Hz (which also is not real) and -1dB around 2.5kHz
Around 0dB at 10kHz (the Tyll rig peak is clearly visible)


Below Golden Ears....
http://en.goldenears.net/en/files/attach/images/254/616/011/3c7a27dace4e4c6da9f3fc979f578d01.png
(direct links are not allowed ?)
Let's not use their green line as reference, but the measured response.
30Hz = 0dB opposite 1kHz. A +3dB hump around 150Hz, 0dB at 2.5kHz (with Pinna) and -1dB at 10kHz
When using their target response (quite similar to mine and Sonarworks's) we get different numbers.
100Hz = 0dB opposite 1kHz. A +2dB hump around 300Hz !, 0dB at 2.5kHz (with Pinna) and -1dB at 10kHz


Next up is Sonarworks, in this case not an averaged plot but an individual one.

50Hz = 0dB ref 1kHz, 150Hz hump = +2dB, 2.5khz = -1.5dB, 10kHz = +3dB and interestingly enough +3dB at 6kHz ?

Of course there are plenty of other plots around.
On SBAF for instance there are loads of measurements from different 'measurebators' but won't post any of their plots here to avoid nasty reactions. But feel free to look around there.

And let's not forget Jude Mansilla (Head-fi)... (gray dashed line, the blue line = HD6XX which is HD650)

50Hz = 0dB ref 1kHz, 150Hz hump = +3.5dB, 2.5kHz = -3dB, 5.5kHz shows a +3dB peak, 10kHz = -7dB but veers up again above it to 0dB (like mine are)

I am convinced that the production spread is very small so all measured differences come from differences in measurement rigs and or compensations applied.
As can be seen in this analysis individual measurements differ several dB's depending on the rigs.
If one were to apply an FR compensation to measure 'flat' within 1dB we would get substantially different sounding headphones as a few dB is quite audible where as when no compensation (based on measurements) would be applied they would all sound the same to the same individual.
The reason people perceive them different (from warm and subdued treble to neutral to bass shy and too forward) is personal preference and not HRTF based IMO. The same reason as why some find the AQ Nighthawk realistic and others find it bloated fat and dark.
Could be seal issues of course but not with the HD650.. I digress.

My point of view here is the following. Basing 'exact' EQ on measurements is flawed simply because of variances in measurement rigs.The HD650 is the most measured headphone in the world and 'exact' compensations aren't recommended when you look at the analysis above.
And remember most are from calibrated and compensated expensive HATS with prof equipment.

What I do see however is a clear correlation between all measurements. In my opinion one should thus not compensate 'exactly' acc. to measurements (as one compensates to the flaws of the rig which are compensated rather than the actual response) BUT must look at the great picture and apply some 'gentle' and average EQ in the area's which need improvement.
Remember, when actively 'compensating' narrow peaks and dips we also get ringing in the audible band.

So IMO a little subbass boost (a few dB) say below 50Hz as that's where the average seems to be (and mine are) and lower the midbass hump (around 150Hz) which all seem to show but in varying levels so around -2dB perhaps with the same bandwidth as all plots show.
I don't think around 2.5kHz nor at 5-6kHz any action should be taken when averaging all plots nor at 10kHz.
All in all that is all that seems needed and exactly what I apply (Analog EQ, not digital).

Now apply this 'knowledge' to all other headphones and realize that the Pinna effect varies (thus between 2kHz and 5kHz) greatly between measurement rigs and different headphones and you could come to the conclusion that only some 'mild' overall compensation and perhaps the largest peaks (sibilance area 6-8kHz) need compensation seems the best way.

Personally, I think Schrodingers method isn't as bad but stands and falls with the SPL at which it is done and is individual as hearing damage is also included.

end of rant... sorry for the excess in words.

 

[andreasmaaan]

Thanks @solderdude, that post was really well-explained

 

[Shikamon]

I second that @andreasmaaan. Really good explanation. Sennheiser make some really good headphones. I kinda miss my HD 558s that had the dead right driver.

Even now I EQ my headphones using noise bands and they sound awesome! Might consider myself getting a better speaker to do my equal loudness with.

Another good advice is to take care of our ears and try to listen to music in safe audible levels as well as cleaning our ears safely from earwax buildup.

 

[pkane]

As a comparison here is my measurement of a recent HD650 (its compensated in the lows)
A horizontal line is 'audible flat' to at around 70-80dB SPL.
In your case the blue line is audible flat.

This is how they sound to me as well (at that volume). Somewhat lacking in the subbass apartment with a hint of 'warmth'(hump around 150Hz) and not needing any correction at all in the mids and treble.
That is indeed why it is universally loved and still today is one of the most neutral headphones.

Hi @solderdude, that's a very impressive analysis! I'll respond to it over a few posts, as I digest this

How did you measure your HD650s? You also mentioned compensation in the low frequencies. What was compensated and what curve did you use for this?

I thought it would be interesting to overlay my in-ear measured response with yours. Trying to match the scales as best as I can in Photoshop. Mine is green/cyan color (I tried to level-match them around 1KHz):

There are obvious differences. Later on I'll post my attempt to measure HD650s outside my ears, for comparison.

 

[solderdude]

Hi @solderdude

How did you measure your HD650s? You also mentioned compensation in the low frequencies. What was compensated and what curve did you use for this?

explained here: https://diyaudioheaven.wordpress.com/headphones/measurements/

scroll down to below the link of all my measurements, there you can find answers to your questions.
Below 400Hz the differences appear to come from my compensation for the lower frequencies.
When I would not do that bass would overlay your measurements mostly.
You solved that by using a 'speaker in room curve'. I solved it by using a hardware low shelve filter in the mic amp.
From 1 to 3 kHz differences I attribute to me not having a Pinna.
From 5kHz to upwards I suspect it has something to do with the chain: mic -> preamp.
I know that the mic amp I designed and built has a bit high noise levels but know for certain it is perfectly flat over the entire range (measured) and is purpose built for the used mic, so no chances of any mismatch or roll-off other than the $ 2.- mic. capsule itself.
As that mic capsule is well documented and even corrected (hardware in the mic amp) for the known peak it has near 16kHz I kind of assume the measurements are kind of accurate to at least 20kHz.

I really should start to look for a Pinna again, maybe find a way to mold my own and then try to make one of silicone or something.
Not buying an 'official' one, way too much money.
Can't use the in-ear method because distortion and spectrum (as well as CSD) measurements will be 'off'.

Have some hum in my rig as well which sometimes shows in FR plots as a small dip or peak at 50Hz.

 

[pkane]

explained here: https://diyaudioheaven.wordpress.com/headphones/measurements/

scroll down to below the link of all my measurements, there you can find answers to your questions.
Below 400Hz the differences appear to come from my compensation for the lower frequencies.
When I would not do that bass would overlay your measurements mostly.
You solved that by using a 'speaker in room curve'. I solved it by using a hardware low shelve filter in the mic amp.
From 1 to 3 kHz differences I attribute to me not having a Pinna.
From 5kHz to upwards I suspect it has something to do with the chain: mic -> preamp.
I know that the mic amp I designed and built has a bit high noise levels but know for certain it is perfectly flat over the entire range (measured) and is purpose built for the used mic, so no chances of any mismatch or roll-off other than the $ 2.- mic. capsule itself.
As that mic capsule is well documented and even corrected (hardware in the mic amp) for the known peak it has near 16kHz I kind of assume the measurements are kind of accurate to at least 20kHz.

I really should start to look for a Pinna again, maybe find a way to mold my own and then try to make one of silicone or something.
Not buying an 'official' one, way too much money.
Can't use the in-ear method because distortion and spectrum (as well as CSD) measurements will be 'off'.

Have some hum in my rig as well which sometimes shows in FR plots as a small dip or peak at 50Hz.

My mic amp and ADC is Behringer UMC22, calibrated and compensated for by REW through a loopback connection. The microphone, on the other hand, is not calibrated, but advertised with a very flat frequency response by the manufacturer.

I have another calibrated measurement mic, but not sure how to use this to properly calibrate the in-ear mic. I'll need some sort of set up where I can position both mics in exactly the same spot in respect to some high-quality transducer, and then compare the results. Have to think more about this. Any suggestions?

You solved that by using a 'speaker in room curve'. I solved it by using a hardware low shelve filter in the mic amp.

Just to be clear, what I posted (and compared to your measurements) is with no DSP or EQ of any kind. There was no room curve, I just placed the Harman line on the same chart to illustrate that the in-ear response of HD650 closely resembles the recommended Harman room curve of -10dB.