How to automatically EQ your headphones to online measurements: a step by step guide

[yatekei]

At this stage you would use the program Sinegen ( http://sinegen.en.lo4d.com/ ) to adjust the high frequency response, since your post-EQ results will almost certainly not match what your ears actually hear. To do this, use an image of whatever target curve you wish to use for reference. Let's say hypothetically we're EQing to diffuse field curve:

After loading your EQ settings in EAPO, sweep through 1000 - 20,000 Hz with Sinegen while listening, listening for obvious peaks and valleys and keeping in mind your reference curve. So, if you're using diffuse field curve, it should steadily increase in volume up to ~3 kHz then steadily decrease in volume after that.

It seems to me the graph ^ that we are using as a reference for the frequency sweep, is just the diffuse field target curve, rather than a perceived loudness curve (for a headphone calibrated to the diffuse field target).

If I'm right, then what is the use of 'keep in mind the reference curve' when doing a loudness sweep in sinegen, if the reference isn't a perceived loudness curve?

You said that 3kHz is meant to be the loudest frequency. Are you using something else entirely to determine that, like an equal loudness contour? Is that what I should be using as a reference instead?

doing that with an equal loudness contour in mind for a gross idea of how we're just less sensitive in the upper and lower range, that can work well indeed after getting used to doing it. but just trying to even things out so the all sweep feels flat, that's not neutral at all.

If you do need to use an equal loudness contour for the frequency sweep, then for example, it would show that 1kHz and 2kHz should be the same perceived loudness, and also there should be a small dip at 1.5kHz. Should I equalise so that there is no dip at 1.5kHz, or should you expect to hear this slight dip on a neutral headphone?

In summary, I'm finding the sinegen frequency sweep very difficult, since I have little idea, or any reference, of how either a harman or diffuse field target (or any kind of neutral target) should sound perceptibly in terms of loudness of pure tones.

 

[TheSonicTruth]

So does anyone have actual numbers for a MDR-7506 ot V6? Does the software in O.P. produce a list of what freq, Q, and gain were used?

 

[toby23]

After losing my calibrated EQ settings after reinstalling Windows, I did not want to go through the manual configuration again and found an alternative, HeSuVi, which contains thousands of calibrated headphone presets. I am now using this, without any surround or filters, and the sound is fantastic, perhaps better than my manual configuration..

https://www.reddit.com/r/headphones/comments/9o2i81/hesuvi_20_released_eq_profiles_for_different/

For those of you who don't have a reddit account, I will quote this post below...

HeSuVi version 2.0 was released on early September and is better than ever. The UI has been reworked and is in my opinion much better now. The most exiting new things are the inclusion of equalization profiles for 1100+ headphone models from my AutoEQ project and a crossfeed. Now all you have to do to get your headphones equalized is to install HeSuVi and select the correct model from Equalizer tab. Some people may not like speaker virtualization (because of HRTF mismatch) so those fellows can try out the crossfeed feature which expands stereo image with headphones without distorting the sound signature.

For the unaware: HeSuVi is a Windows program for surround virtualization on headphones but also has equalizer and now a crossfeed as well. Surround virtualization works on any stereo headphones and can have major improvement especially on games and movies. With personalized measurements this technology can make it almost impossible to distinguish between listening speakers and headphone. More about this later.

 

[yatekei]

+1 for HeSuvi. I use it along with Peace GUI, in the Equaliser APO config editor.

The surround sound virtualisation and huge number of preset EQ's HeSuvi offers is great, but in my case, for my in-ears and m50x headphones, the presets seem to have a lot of error. +-20db dips and peaks, different on each headphone.

For instance, I have a very sharp peak from 3900-4500hz on my m50x. By EQing, a roughly -20dB drop is required to retain equal loudness with surrounding frequencies (I aimed for smooth equal loudness without any major dips or peaks from 3000Hz to 5000Hz).

The HeSuvi EQ presets don't come close:
- The 'HP' and 'IF' EQ presets both have an approximately +4dB rise from 3kHz to 4kHz.
- The 'o90' preset has a -5dB dip from 3kHz to 4kHz.

Given that there are sources of error in your ears, headphones, and their measurements... Using an EQ to a neutral target curve with online frequency response measurements can be insufficient.
- Our ears have a unique HRTF, which can vary significantly.
- Your headphones might be faulty / unique, so their frequency response is different to online measurements.
- The measurements from different sources can be measured unreliably or inaccurately.

Which I think is why most EQ guides I've seen (including this one) recommend also doing a tone frequency sweep as well, but I am uncertain on how exactly to do that.

I found an EQ method which elaborates on loudness testing over a frequency sweep using a generic equal loudness contour. Ideally you should use an equal loudness contour measured for your own ears using a perfectly neutral speaker system).
https://www.head-fi.org/threads/how...dphones-advanced-tutorial-in-progress.615417/

 

[EQGuy]

HeSuvi looks super convenient! Be aware it appears, to me, to auto-EQ to something akin to the Harman curve. That's a bass-boosted curve. Some may not like this, but I suspect most will.

Yatekei, EQing to an equal loudness contour is not something anyone should be doing. Equal loudness contours and proper headphone target curves are two radically different things. Also, I didn't really intend this to be a guide to thoroughly understanding various target curves. One could write a book on that subject. Sorry, but I did kind of write the guide for people with a basic understanding of target curves like diffuse field, Harman target, etc. You'll need to do a bit more research on target curves; you'll need to understand the concept of curves like that as well as the curves' general sound (diffuse field will probably be subjectively perceived as bright; Harman target will probably be subjectively perceived as neutral to bassy) to really understand the tutorial in general.

 

[castleofargh]

what @yatekei mentioned is a fairly old method (that I used a lot when I had nothing better), where you use your own equal loudness contour to set a reference between 2 headphones. it was for people who didn't even have the frequency response of their headphones. or for people who wished to try and get let's say a headphone with the frequency response you perceive when using your speakers. the obvious benefit is that you sort of calibrate everything with your own hearing, making sure your HRTF is involved the way it should(at least the basic frequency response part of it at one fixed location).

this uses the same idea, for a slightly different purpose https://www.head-fi.org/threads/frontal-sound-and-correct-frequency-response-with-eq-only.853443/ Griesinger's purpose is for us to setup a frequency response that will ensure that a mono sound will feel like it's in front of us, instead of up or below like is often the case with headphones of various frequency responses. it's a pretty cool idea IMO, but for that to reach some idea of getting a flat headphone signature, it would require better than the cheap tweeter thingy he used in the video. but that idea can certainly be used on a well EQed speaker in near field configuration in front of us. I played around with those a lot and even got the little program that Mr Griesinger was nice enough to share. at the time it wasn't all that good, as it was very(very!) easy to end up clipping the signal while EQuing for the calibration and it required some getting used to so that the gain would remain fine. I'm talking about some times back, so maybe he now has a more functional version?

nowadays I would suggest to save a lot of time and test tone annoyance by purchasing the cheapest binaural microphones you can find(the sort you put in your own ears, not some dummy heads) and simply measure the FR of the sound source you wish to replicate into your headphone, then measure your headphone, and do the math to get the final EQ. there is a cost, last I checked most were close to a hundred bucks, so there lies the real question. is it worth it? it's way faster for sure.

anyway none of those methods claim to offer neutral response(not that anybody can for headphones IMO). they offer to copy the FR of something onto something else.

 

[yatekei]

I think I understand target curves and your tutorial, but only in theory.
https://www.innerfidelity.com/content/headphone-measurements-explained-frequency-response-part-one

But I certainly don't know how they are supposed to sound in terms of loudness (except relatively to each other. I can see how diffuse must be brighter by some degree, and that harman will have a bit more bass)

So when you said you should keep in mind the diffuse field reference when doing the frequency sweep, I'm supposed to already know how that sounds in terms of perceived loudness? Do you have a reference or just an extremely vague idea?

'EQing to an equal loudness contour is not something anyone should be doing'

Is there no relation between the perceived loudness of a neutral headphone (i.e. if it is calibrated to the diffuse or harman target) and the equal loudness contour then? In other words, if you get a 'neutral' headphone and do a frequency sweep over it, you're saying it won't correspond in loudness with an equal loudness contour? Surely, just like the equal loudness contour, at least the sub bass and high treble loudness would be attenuated, and you mentioned that 3kHz is the loudest frequency as well.

 

[EQGuy]

Castleofargh, Griesinger published a file named Binaural_hearing_and_headphones.ppt (you should be able to find with Google and/or Internet Archive) which contains instructions for building affordable in-ear microphones that rest at the eardrum. These would be absolutely ideal for EQing headphones. The only problem is they use a Radio Shack part which is obviously now unavailable. If anyone can suggest a similar affordable mic capsule for that purpose, please do so. Also, for those with 4 digits to drop comfortably, Etymotic makes an inner ear probe mic system.

Is there no relation between the perceived loudness of a neutral headphone (i.e. if it is calibrated to the diffuse or harman target) and the equal loudness contour then?

Correct. I am not aware of any headphone target curves which utilize equal loudness contours in any way.

 

[TheSonicTruth]

I think I understand target curves and your tutorial, but only in theory.
https://www.innerfidelity.com/content/headphone-measurements-explained-frequency-response-part-one

But I certainly don't know how they are supposed to sound in terms of loudness (except relatively to each other. I can see how diffuse must be brighter by some degree, and that harman will have a bit more bass)

So when you said you should keep in mind the diffuse field reference when doing the frequency sweep, I'm supposed to already know how that sounds in terms of perceived loudness? Do you have a reference or just an extremely vague idea?

'EQing to an equal loudness contour is not something anyone should be doing'

Is there no relation between the perceived loudness of a neutral headphone (i.e. if it is calibrated to the diffuse or harman target) and the equal loudness contour then? In other words, if you get a 'neutral' headphone and do a frequency sweep over it, you're saying it won't correspond in loudness with an equal loudness contour? Surely, just like the equal loudness contour*ELC*, at least the sub bass and high treble loudness would be attenuated, and you mentioned that 3kHz is the loudest frequency as well.

Actually, the ELC(see my insert in your last paragraph), or Fletcher Munson curve, for dinosaurs like myself(!) indicates how much louder something at, IE: 100Hz would have to be to sound as loud as something in the upper midrange(2-4kHz). So no, you wouldn't want to apply an EQ curve that is complementary to the ELC, at any Phons level. You would just be exaggerating the average human hearing response we were all born with.

For fun, I did replicate a 80phons ELC with the parametric Equalizer app by AudioForge, and I must admit it was a pleasure to listen to music with at moderate to background volume levels. In other words, it let me hear more of the music at low volume settings, especially the bottom end and the top, via both headphones and speakers. But again I did say 'for fun', and the result will be anything but 'flat'. Just a precaution.

 

[TheSonicTruth]

After losing my calibrated EQ settings after reinstalling Windows, I did not want to go through the manual configuration again and found an alternative, HeSuVi, which contains thousands of calibrated headphone presets. I am now using this, without any surround or filters, and the sound is fantastic, perhaps better than my manual configuration..

https://www.reddit.com/r/headphones/comments/9o2i81/hesuvi_20_released_eq_profiles_for_different/

For those of you who don't have a reddit account, I will quote this post below...

HeSuVi version 2.0 was released on early September and is better than ever. The UI has been reworked and is in my opinion much better now. The most exiting new things are the inclusion of equalization profiles for 1100+ headphone models from my AutoEQ project and a crossfeed. Now all you have to do to get your headphones equalized is to install HeSuVi and select the correct model from Equalizer tab. Some people may not like speaker virtualization (because of HRTF mismatch) so those fellows can try out the crossfeed feature which expands stereo image with headphones without distorting the sound signature.

For the unaware: HeSuVi is a Windows program for surround virtualization on headphones but also has equalizer and now a crossfeed as well. Surround virtualization works on any stereo headphones and can have major improvement especially on games and movies. With personalized measurements this technology can make it almost impossible to distinguish between listening speakers and headphone. More about this later.

So I implemented *most of* the following correction for my MDR-7506:


Using my aforementioned AudioForge Equalizer app on my iPod Touch and iPhone. It allows adjustments up to seven frequency points, as opposed to the ten adjustments suggested in the image I attached.

My observations? What I heard was definitely different, but I don't know if correct.

And I can say beyond a doubt that what I heard was not targeting any 'Harman Target Curve' I ever replicated. If anything, it was targeting something *close* to flat, but definitely not flat to my own ears. In fact, I omitted the massive bottom cut at 51Hz, Q=0.48, partly because my own hearing tends to be 'top-heavier' and bass-shier than most, then rolling off sharply above 12-14kHz. Most renderings I have seen of the Harman curve seem to apply a huge +5dB bottom shelf up to 100-200Hz, both target and achieved. I consider the MDR-7506 sub-100Hz presence to be just right for my hearing, not boomy, not too shy, just - there. My goal is Flat, and any Harman target curves stored in my EQ app have been deleted - that thing is just about UNLISTENABLE over any of my headphone collection(including a DT-880 pro and SRH-440). I do not get what the goal of that curve is.

Again, because I am hyper-sensitive to anything upper-mid to low-treble, that positive adjustment at 5.989kHz really stuck out during my half hour audition with the curve implemented. I may reduce the stratospheric boost at that frequency down 1 or 2 dB.

I checked the resulting curve as displayed on my devices against several published FR curves for Sony MDR-7506, and it indeed seemed to very closely compliment most of those published curves, to within just a few Hz and 1-2dB!

I will continue listening with this curve for the next few days and advise of my longer term impressions.

And thank you, toby23, for providing the information!

 

[yatekei]

@TheSonicTruth When I say equalise using an equal loudness contour, I don't mean make every tone the same loudness. I mean use the ECL to ensure that the loudness of tones follows the same curve, to ensure that 100Hz doesn't have the same loudness as 2-4kHz. And 3kHz would always be the loudest tone over the whole frequency range regardless of volume level.

Correct. I am not aware of any headphone target curves which utilize equal loudness contours in any way.

I agree that nobody appears to derive any target curves by the use of equal loudness contours.

Ok I'll be more specific to your tutorial: When you do the frequency sweep, all you can test is perceived loudness. So if, as you suggested, 3kHz is supposed to be the loudest perceived frequency, what are you basing that on, if not an equal loudness contour? I'm still not sure what the 'general sound' actually is meant to be across the frequency spectrum.

According to my suggestion of using the ECL for the frequency sweep, I would suggest that 1kHz and 2kHz should be the same loudness at all volume levels, and that should I expect a small reduction in perceived volume between them at 1.5kHz. Is that false then?

 

[EQGuy]

what are you basing that on

The human ear will perceive ~3 kHz the loudest because of the ear canal+concha system's one quarter wavelength resonance. This resonance is effectively built into all of the various headphone target curves, such as diffuse field, Harman, etc.

I'm still not sure what the 'general sound' actually is meant to be across the frequency spectrum.

The easiest way to understand how the loudness of your sine sweeping should proceed through the frequencies is to sweep through the frequencies while looking at the visual representation (graph) of a target curve like diffuse field, Harman, etc.

 

[TheSonicTruth]

@TheSonicTruth When I say equalise using an equal loudness contour, I don't mean make every tone the same loudness. I mean use the ECL to ensure that the loudness of tones follows the same curve, to ensure that 100Hz doesn't have the same loudness as 2-4kHz. And 3kHz would always be the loudest tone over the whole frequency range regardless of volume level.



I agree that nobody appears to derive any target curves by the use of equal loudness contours.

Ok I'll be more specific to your tutorial: When you do the frequency sweep, all you can test is perceived loudness. So if, as you suggested, 3kHz is supposed to be the loudest perceived frequency, what are you basing that on, if not an equal loudness contour? I'm still not sure what the 'general sound' actually is meant to be across the frequency spectrum.

According to my suggestion of using the ECL for the frequency sweep, I would suggest that 1kHz and 2kHz should be the same loudness at all volume levels, and that should I expect a small reduction in perceived volume between them at 1.5kHz. Is that false then?

" When I say equalise using an
equal loudness contour, I don't mean make every tone
the same loudness. I mean use the ECL to ensure that
the loudness of tones follows the same curve, to ensure
that 100Hz doesn't have the same loudness as 2-4kHz."

But what you said in the second sentence already happens in human hearing!

Again: The purpose of the ELC is to indicate how loud, in SPL, all tones would have to be to sound as loud, to the average listener, as each other. Do your self a quick favor: Copy a typical published ELC into Power Point or another drawing/illustration program. There's one on Wikipedia. Trace it with a thick line, then flip that tracing upside down. Voila! You just drew yourself a typical (healthy, young to middle-aged) person's actual hearing response.

Print it out and tack to your wall, or save to your devices.

By that exercise you should now understand that the ELC is just a compensatory, complimentary curve of the average person's hearing response. Sum the two curves and you get:
20Hz________________________________________________20kHz. (flat, that is)

 

[yatekei]

The human ear will perceive ~3 kHz the loudest because of the ear canal+concha system's one quarter wavelength resonance. This resonance is effectively built into all of the various headphone target curves, such as diffuse field, Harman, etc.
The easiest way to understand how the loudness of your sine sweeping should proceed through the frequencies is to sweep through the frequencies while looking at the visual representation (graph) of a target curve like diffuse field, Harman, etc.

So the very best representation of perceived loudness of a diffuse or harman target curve, are the target curves themselves?

I mean, when the SPL changes at a given frequency on a harman target curve, what is the change in loudness of the tone?

To answer that, I would immediately want to use an equal loudness contour to answer the question.

So, to determine what neutral sounds like, it seems to me like it would involve use of the equal loudness contour.

"
When I say equalise using an
equal loudness contour, I don't mean make every tone
the same loudness. I mean use the ECL to ensure that
the loudness of tones follows the same curve, to ensure
that 100Hz doesn't have the same loudness as 2-4kHz."

But what you said in the second sentence already happens in human hearing!

If 100Hz is louder than 1Khz after attenuation has already occurred in our human hearing, you would know it's too loud. Or if you could hear 16kHz as loud as 1000Hz out of your headphones / speakers, then you can be certain you've got a very bad sound system. Because your natural hearing highly attenuates 16kHz.

 

[EQGuy]

So the very best representation of perceived loudness of a diffuse or harman target curve, are the target curves themselves?

Well when I said that I was referring to the process of sweeping through the upper frequencies with Sinegen for peak removal. Basically, if you take a look at the diffuse field curve I posted in the OP, you'll notice that it peaks at ~3 kHz and then gradually decreases in volume as the frequency rises from there. So you basically look at the target curve while sweeping through the frequencies, see that it peaks at ~3 kHz, then decreases in volume smoothly without peaks from there, so you basically try to approximate that with EQ and Sinegen reference. Also, remember not to EQ out the three quarter wavelength resonance, as that's supposed to be there, as previously explained.