[Duncan]

Oh... I see

Sorry Joe, lol... yeah, I now see that you quoted my thread title, I originally thought you meant that I'd EQ'd my etymotics... and I was like... erm no??

 

[anticowboyism]

I'm probably following in the footsteps of someone's failed eq, maybe your first one...

But here's my plan so far:

Take the AKG's FR from HeadRoom (not sure which now that you mention it) and invert it so it will measure flat. Add the inverse of the HRTF contour to take away the "headphone" sound. Then add the inverse of the Fletcher-Munson so it sounds flat to my ear. I will probably still have some tweaking to do after that, seeing how my ears are probably not the same as the F-M testers.

The Ron Cole EQ is on one of the pages you linked to above. It is to compensate for listening on headphones what is intended for speakers. Basically the same as the HRTF contour in principle, but a competely different graph, which makes it harder to choose which is correct.

BTW, is there any better software EQ than SuperEQ for winamp? I like it, but it doesn't have the resolution needed for this kind of tweaking. I need at least a 1/3 octave eq. The parametric on it looks promising, but I don't know what the slopes are (can they be adjusted?) or what the meanings of the boundary frequencies are. Are they where the graph crosses zero?

Is that true about how HeadRoom normalizes the FR graphs? That's kind of disappointing. I made an inverse to the smoothed normaled of my phones. It sounds all right. ? Very different. I'm not sure which I like better...

It definitely has the bass I've been looking for. But the mids seemed to mellow out alot. I don't know if I like that, or if thats more accurate. I guess I should go listen to these CDs on the Genelecs at the studio to see how they should sound...

 

[Joe Bloggs]

I go out for a few days and see what a mess things become

Quote:

Take the AKG's FR from HeadRoom (not sure which now that you mention it) and invert it so it will measure flat. Add the inverse of the HRTF contour to take away the "headphone" sound. Then add the inverse of the Fletcher-Munson so it sounds flat to my ear.

...
The HRTF contour are meant to be directly added to the FR of a pair of headphone drivers with flat response. That is, the HRTF is the difference between the ear response to a pair of flat-FR earphones and a pair of flat-FR loudspeakers.

So provided that you've managed to EQ your phones to flat (as measured with a acoustic coupler), you should add the HRTF directly, not the inverse!

Since none of the HeadRoom graphs are measurements against flat response, none of the graphs can be used for this purpose.

What kind of F-M (Fletcher-Munson) curve are you looking at? If it's one of these

with a whole stack of contours...

These contours are supposed to match loudness as perceived by the average person at different frequencies with the actual dB SPL (Sound Pressure Level). As an example, this graph states that for the average person to perceive an 80dB 1kHz tone and a very low ton, ~25Hz at the same loudness (they sound as loud as each other), the 25Hz tone has to be at 90dB SPL instead of 80dB.

Sooo, suppose you take one of these lines and add them to your EQ... (for example the 80dB line) And supposing your speakers have perfect flat response or your headphones are perfectly diffuse-field equalized (e.g. EQed to measure flat on acoustic coupler + HRTF EQ)...

After all that, when you play a constant amplitude test tone sweep that goes from 20Hz to 20000Hz, the average ear is supposed to hear the whole sweep at the same perceived loudness...

IF you adjust the volume knob such that at the 1kHz part of the tone sweep the speakers/headphones are playing at 80dB SPL.

None of this, however, has anything to do with an ideal frequency response for music playback.

That's because records are mastered to sound best on a system that MEASURES flat, not one that SOUNDS flat. A system EQed to SOUND flat to your ears with a tone sweep will have incredibly loud bass, little mid-treble and shrieking high treble.

Applying the INVERSE of the F-M curve requires even less discussion than the, um, inverse of the inverse. It serves no purpose whatsoever.

UNLESS...

You EQed the drivers so that all parts of the test tone sweep are of equal perceived loudness to start with.

Since the F-M curves specify the EQ changes required to turn a speaker in front of you from MEASURING flat to SOUNDING flat to the average pair of ears at a given volume level (xdB @ 1kHz)...

IF A PAIR OF SPEAKERS, HEADPHONE DRIVERS OR INDEED ANYTHING THAT CAN MAKE SOUND, *SOUNDS* FLAT TO YOUR EARS AT A GIVEN VOLUME LEVEL (xdB @ 1kHz), APPLYING THE INVERSE OF THE APPROPRIATE F-M CURVE (the xdB contour) CAN THEORETICALLY CHANGE THE RESPONSE OF THE SOUND MAKING DEVICE FROM *SOUNDING* FLAT TO

SOUNDING LIKE A SPEAKER PLAYING IN FRONT OF YOU THAT MEASURES FLAT

What that means is, if the sound-making device IS in fact a speaker playing in front of you, it would be EQed to measure flat. If it's a headphone, HRTF would be automatically applied to make the phones sound like a speaker playing in front of you that measures flat. (that's because you effectively applied the HRTF using your ears when you were EQing the phones to SOUND flat in the first step)

Sounds good doesn't it?

Well I tried it. It didn't work for me, because the PUBLIC F-M curves specify the response of the AVERAGE ear. If your ears are not bog-standard, your own F-M curve will be much different from any of the numerous public F-M curves.

If you've read so far, you're probably confused as hell and waiting for a summary, so here it is:

If you don't have measuring instruments, EQ may be a waste of time, but you can try EQing your phones to SOUND flat to your ears across all frequencies at a certain volume level and apply the inverse of the appropriate F-M curve (e.g. if you listened loud, at an estimated 80dB SPL @ 1kHz, use the 80dB curve)

OR you take a listen to the best phones on earth, EQ them to flat-SOUNDING at a given volume (again, say 80dB SPL @1kHz) they should have MEASURED flat to start with--that's a different thing, remember) note down the EQ adjustments you made--call it alpha

Then EQ your own phones to flat-SOUNDING at the same given volume (80dB SPL @1kHz). Then apply the inverse of alpha. You should now have the same response as that great pair of phones unEQed.

Joe

 

[j-curve]

Quote:

anticowboyism: Is that true about how HeadRoom normalizes the FR graphs?

I seem to recall something on their website along those lines, that all their graphs are normalised against their "best 10". However, looking at the graphs I got the feeling that the simple frequency response graph was pretty raw, as measured on their dummy head. If not, what are the "normalised" graphs normalised against?

Looking at the basic HD580 graph, the low end drops off as you'd expect from a headphone, despite Senn's claim of -3dB at 16Hz using an acoustic coupler. But the normalised graph shows that the low end of the HD580 is actually flat or slightly up relative to the other nine headphones. That's my interpretation anyway... could be wrong. And it doesn't change what Mr Bloggs is saying about being dummy head measurements as opposed to acoustic coupler measurements.

BTW Joe, I wanted to ask you about how you apply an EQ curve via your filter software. As you know, on a constant-Q equalizer, boosting a band by 6dB adds about 2dB to the adjacent band and 1dB to the next one as well. The problem being that if you have a curve you want to match then you won't get the desired result by simply dialling up the curve on the graphic. For example, if you want
dB (target) 0 6 9 12 0 in a series of bands then you have to dial up something more like
dB ( set ) -2 4 4 12 -4.
Hardly intuitive! In this example, overlooking this correction puts all these bands 4-5dB above their targets.
So my nitpicking question of the day is: do you (or does your software) allow for this?
Cheers.

 

[Joe Bloggs]

Quote:

As you know, on a constant-Q equalizer, boosting a band by 6dB adds about 2dB to the adjacent band and 1dB to the next one as well.

You overestimated my knowledge

No I didn't know, nor do I know what to do about it either

Well, unless I can get specs of the software EQ from Naoki Shibata da man himself, I don't see myself changing my EQ--I'm too lazy to second-guess the EQ behaviour and I've no experience in this

On a brighter note, thanks for pointing this out to me!

Looks like there may be room for improvement in my EX70->Ety EQ yet

What is the standard behaviour of such an EQ? Can you write down a table of sorts?
6 dB boost: 1 2 6 2 1
3 dB boost: 0 1 3 1 0?

Hmm, but I wonder whether this applies to this digital EQ, since Shibata said that an overall EQ line is sort of plotted in the program before it is applied to the incoming stream.

And how about for cut? All the settings in the EQ are cuts actually, not boost--to prevent digital clipping.

Well it seems there are always new things to learn

I just hope anticowboyism comes back to read this to avoid making even greater errors than I am (possibly) making.

Joe

P.S. The 'normalized frequency response' uses the average of the '10 best sounding headphones' according to HeadRoom as the yardstick. The theory (I suppose) is that the different shortcomings of the 10 phones would cancel each other out and the yardstick would then represent a very well-balanced pair of phones. But there could be commong shortcomings that all headphones share and extreme low bas may well be one of them.

 

[j-curve]

Quote:

Can you write down a table of sorts? 6 dB boost: 1 2 6 2 1

You just did, and that's all you need to know. Label five adjacent band frequencies "a" to "e". If your graphic is set and you want to know the approximate total gain you're getting at c, your best estimate is:-
Gain_c = Slider_c + 1/3 * (Slider_b + Slider_d) + 1/6 * (Slider_a + Slider_e)
(The sliders will be + or - for boosts and cuts).

In the opposite direction (ie. setting the graphic to match a desired curve), I'm not sure there's an easy answer since the sliders depend on each other, but there are various ways you can tackle it. One is to start at one end of the curve and set the first slider to the target gain with no correction. Then as you proceed to each new slider you have to make small adjustments to the previous two. It doesn't come out exactly right in one go but unless you're doing crazy boosts or cuts

it will be close. Or you can just dial up the curve as normal and then make skillful adjustments based on the 1-2-6-2-1 shape. In general this will mean reducing the amount of boost or cut when all the sliders are in one direction, but actually increasing the amount of boost and cut wherever the curve crosses zero.

Joe, you usually do almost all cuts, right? I think that would put you at risk of over-cutting due to the 1-2-6-2-1. I'm wondering if that's what happened when you "adjusted"

the HD580(?) [I'm not going to try and find that post again since you're such a controversial dude that your threads go on forever!]

Well, I'm not a graphic guru by any means. It's just that I've been getting interested in this stuff and have been thinking of doing some tweaking myself, on the fly of course, no flaming CD's for me. Not wanting to have to tweak separate left and right sliders, I thought what I wanted was this:-

When I saw the price though, I "re-evaluated" the inconvenience of dual sliders and got tempted by one of these:-

(click for larger image)

But not being able to store and retrieve curves in a flash felt like such an endless knob-twiddle that I found myself considering the digital route. Hell, most of the stuff I listen to has already been DACked and unDACked 2-3 times so why not DACk it again for good measure? The same company makes a bells-and-whistles spectrum analyser/graphic/room-tweaking monster:-

(click for larger image)
but there are various issues involved. One is price again, although for what it does it could hardly be called expensive. For a little extra you can even avoid the extra ADC/DAC stage if you purchase their optional digital interface, although you then get involved with sampling rate conversion since the internals operate at 46.875kHz as opposed to CD's 44.1kHz. Another issue is reliability problems reported around the net which seem to plague this model. (At least the problems start on day one, not day one-year-plus-one). You might want a wheelbarrow to move it around the house, too.

Then I noticed this little beastie:-

(click for larger image)
(product info here)

What they call Feedback Destroyer, I call 2 x 12-band parametric equalizer. [

Doh! I was hoping to get through this post without using the "e" word!

] You can tune the channels separately or link them for stereo, with 10 memories for your favourite curves. Now, parametric may seem user unfriendly and hard to adjust but I think the greatest sticking point has been the feeling of operating in the dark, especially when you need high Q. An analog frequency knob can be hard to tune accurately and when you get it tuned you have no idea of the exact settings anyway. Digital changes all that.

There is also free filter design software for the DSP1100P (20-bit predecessor to the 24-bit DSP1124P pictured above) which runs on Windows and displays total response curves. Check out this software if you get a chance. One quirk, these guys specify bandwidth in 1/60ths of an octave, so to simulate a 1/3rd octave graphic (well, 12 bands of one) you have to set the bandwidth to 20. Not knowing if 12 parametrics would match the flexibility of a 31-band graphic, I used their software and tried to draw a response matching the HD580 curve at Headroom, in a free-for-all parametric versus simulated 1/3rd octave graphic shootout. My conclusion is that the parametric is far superior, giving a closer match with lesser boost and cut.

Admittedly parametric is pretty much a geeks-only zone but the software really helps. Supposedly you can download your perfect filter design to the unit via a MIDI interface, to save you from having to battle with the clunky front panel.

Attached to this post is a parameter file for the software, with some numbers in it especially for you.

PS: It's a .fbx file, so you'll have to rename it since Head-Fi doesn't like the funny extension.

 

[Joe Bloggs]

Quote:

Joe, you usually do almost all cuts, right? I think that would put you at risk of over-cutting due to the 1-2-6-2-1. I'm wondering if that's what happened when you "adjusted" the HD580(?) [I'm not going to try and find that post again since you're such a controversial dude that your threads go on forever!]

Hmm... I would think that even if one band has an effect on the next the effect would be based on the *relative* difference between two bands, right?

I mean, if you have an EQ with all bands set to -6dB (let's not worry about why you'd do such a thing in the first place

), it would just come out as -6dB on all bands, right? If you go calculating the effect on adjacent sliders by the absolute amount of cut, you'd conclude that you get -9dB on all bands, which just doesn't seem right

If the effect is calculated from relative difference, the only effect of not accounting for it is that the peaks and dips are not as sharp as you want them to be--there should not be any effect on overall gain.

Nice software you have there! If only I can use it

BTW, what is the attached EQ for? It doesn't seem to correspond to anything I currently use

 

[j-curve]

I did an experiment with the simulated graphic today, dialling up 6 6 6 on three sliders. [What can I say,

EQ

is evil.

] Ignoring the side bands, the response came out at 9 10 9 as predicted by the 1-2-6-2-1 effect, smoothly joined. Then I tried trading off smoothness for more accuracy in the gain. By reducing the bandwidth of the filters down to 60% of the original value I got 7 8 7, with dips down to 5 inbetween. In other words I had changed 1-2-6-2-1 into 0-1-6-1-0, but paid for it with more ripple in the response and a wobbly phase graph. [Do we care about phase?

]

Then it occurred to me that just because a 31-band graphic has sliders spaced every third of an octave doesn't mean the actual filters have the same bandwidth. Manufacturers don't quote their figures but a quick search scored this fantastic article called Equalizer Inequality. Fascinating but rather long so I'll just quote this gem:-

Quote:

BANDWIDTH: A DECISIVE PARAMETER Graphic equalizers are "constant-Q" devices, where each filter has the same percentage of an octave bandwidth as the next. Few people realize, however, that the standard for graphic EQs refers only to the spacing of the center frequencies, and not the bandwidth of the individual filters. Depending on the model, the width of the filters can vary from a full octave to 11/46 octave. Two units with identical visual settings can sound radically different due to the interaction of adjacent filters.

Bingo! Now 11/46ths is 72% of the textbook design (1/3rd of an octave) so even the most aggressively curve-fitting manufacturer won't achieve 0-1-6-1-0. As a result, slider settings do not show gain curves, gain curves cannot be dialled directly into sliders, and you won't get what you expect if you add one set of slider values to another set of sliders or to a curve.

Quote:

Joe Bloggs: ...even if one band has an effect on the next the effect would be based on the *relative* difference between two bands, right? I mean, if you have an EQ with all bands set to -6dB..., it would just come out as -6dB on all bands, right? If you go calculating the effect on adjacent sliders by the absolute amount of cut, you'd conclude that you get -9dB on all bands, which just doesn't seem right...

Nope, it's absolutes, and it would be -12dB, not -9dB. Try it out! If you don't have any measuring gear then maybe the Spectrogram 5.0 spectrum analysis freeware would come in handy. Not only does it have dB readouts but when fed with filtered white noise it will show you exactly what your EQ is doing. A bit of a fiddle though since I think it only works on your soundcard's line/mic input.

I know you want to avoid digital distortion at all costs but as far as the side-effects (gain ripple, phase wobble & 1-2-6-2-1) of EQ you would be better off to have an average slider value of zero. You would have to be unlucky to get a full volume passage of music in your frequency band of greatest boost, wouldn't you? Maybe you could reduce the volume by a few dB before the EQ?

BTW, apologies if it appears I'm taking pontificating pot-shots at your methodology. You certainly can't go wrong when tuning the graphic by ear and you obviously have good ears to be able to do that.

If what I'm suggesting about graphic EQ's is right though, the curve you get is not what the sliders say. You have to calculate the curve from the sliders (easy), do your addition and subtraction of curves as usual and then calculate the slider settings you need to achieve the end result (not quite as easy). Once again, sorry for the rant. Hopefully the day that I add experience to my air-guitar theories is not far off!


PS: The attachment should have had the bottom 12 bands of the Shibata EQ programmed in as a simulated 1/2-octave graphic EQ. Once you load the file you might have to bring up the smaller window with all the rows of text in it, choose "Program 2" from the "Memory" pull-down menu and then click on the "Load" button to transfer the settings to the graph window. (Did all that make sense?) There is also a "Save" button next to the Load button, for saving your tweaks into the Program slots. Have Fun.

 

[a1leyez0nm3]

why not just run a FR scan audio file, and make it so that all the frequencies sound at the same level... thats what i did to start with...

 

[Joe Bloggs]

Hmm... I may have to ask Shibata about this, his EQ may indeed work differently... For one thing, the GUI only lends itself to cut, not gain (ie. the max setting for each slider is zero, everything else is -ve) And for another, some of my music have indeed clipped even with the highest slider set at -1dB and everything else lower than it (the two next to it are -2 and -6)

(How is it possible to get clipping with everything set to -ve? Because of the nature of the digital calculations involved, passing a waveform through a straight EQ may result in some peaks higher than the original. If this effect is great enough you can get clipping even with a little cut)

But going by your reasoning, the highest point of my EQ should already be -3 or -4, and I find it quite hard to believe that it the EQ could make peaks stand out 3 or 4dB higher than they should!

BTW, the Shibata EQ uses FIR filters, so should be phase-linear.

But there are problems with pre-echo, just like mp3s although to a smaller extent.

 

[j-curve]

The world looks to have a bit of curvature to it and I'm beginning to doubt if it's flat at all...

I feel as if my air-guitar has been confiscated.

I agree with you that the numbers just don't seem to stack up. Why should an EQ be so out of whack with its sliders? I set out to understand why and ended up building a spreadsheet to do all the number crunching. That way I could get a handle on what exactly is going on. It was a bit of a saga but here's what I've got so far:-

The 12-band parametric EQ software models a chain of filters in series. Hence the effects of the separate stages are effectively multiplied together, allowing you to simply add the dB gain and phase curves together to get the total response. Provided that's how the DSP in the actual product works, no problem. Analog graphic EQ's are rarely made this way due to the noise buildup in the cascaded stages. With digital I would have thought that the time latency of such a design would make it impractical, but evidently it doesn't.

By contrast, a graphic EQ usually consists of a bank of bandpass filters in parallel, whose outputs are weighted by the sliders and summed together. This addition, combined with mild phase shifts across the center frequency of each band, means that there is some cancellation between the signals which allows you typically to achieve 0-1-6-1-0 when combining adjacent bands!!

However, don't think I would relinquish the 1-2-6-2-1 theory so easily! When you move a single slider the effect is still present. Calculated accurately it is actually 0.4, 1.9, 6, 1.9, 0.4. [To be fair then, perhaps we should rename it the 2-6-2 effect?] It is a real phenomenon for both graphic and parametric equalizers when the bandwidth matches the spacing between the centre frequencies. Which bandwidth though? There's only one, isn't there? Well, now things get really sticky...

In most cases constant-Q is strictly speaking a misnomer(!) Manufacturers tout their constant-Q designs by comparing them to the "conventional" (ancient) RLC or "gyrator" design whose bandwidth was very wide for small amounts of boost & cut, and very narrow for larger settings. Problem solved then? Not quite. The modern design certainly has a constant-Q part - its bank of 2nd order Butterworth bandpass filters. But when you add a weighted bandpass response to the original signal the resulting EQ curve has tails which approach but never quite reach zero, making it hard to establish a peak-minus-3dB point for small boost or gain. On a 1/3rd octave graphic, a 3.5dB boost still comes out with a bandwidth of an octave, and a 6dB boost (or cut) has a bandwidth 41% larger than it should (which comes out nearly 1/2 an octave). This (and the 2-6-2 effect) can be reduced by using narrower Butterworth filters, a tradeoff against ripple and phase wobble as mentioned before. For the 6dB case, the appropriate reduction in bandwidth would be to 1/1.41 = 71%, explaining where the 11/46ths octave filters came from - an EQ whose bandwidth was corrected for 6dB boosts & cuts. [Incidentally, that bandwidth correction also achieves 1-X-1 (in the case of two adjacent sliders at X) for X up to at least 48dB(!).] In any case, to truly achieve constant-Q for any boost or cut would require boost-dependent bandwidth adjustment of the Butterworth filters. With the complexity and cost involved I doubt any analog product available does this.


I wasn't having much luck modelling the DSP1124P's parametric filters until I realized that the bandwidths in the DSP1124P software are being tweaked depending on the amount of boost or cut. This seems inconsistent with the fundamental premise of a parametric equalizer - that the user gets to choose the parameters! The tweaks look like this:-
dBbandwidth correction
-489%
-3617%
-2433%
+/-1654%
+/-1267%
+/-886%
+/-696%
+/-4107%
Edit (4th March 2003): The above numbers were derived from the free-download software for the DSP1100P and do not concur with my measurements of the actual response curves of the DSP1124P(!), which are as follows:-
dBbandwidth correction
-4817%
-3235%
-2455%
+/-1682%
+/-12100%
+/-8110%
+/-6127%
+/-4150%

This makes an easy curve-fit, but the whole concept seems "highly illogical, Captain". A one-octave filter secretly becomes a half-octave filter if you cut by 16dB (actually, about 26dB). Perhaps the 2-6-2 effect can go critical due to the compounding effect of series-linked filters. Or maybe narrow notches take first priority since the gizmo is primarily intended as a Feedback Destroyer

Bottom line is that I must withdraw my recommendation of the filter software for modelling generic analog graphic EQ's, due to the fact that the topology of the two equalizer types is different and thus they give different results when multiple sliders are moved. You could model single slider effects but having to reverse out the bandwidth correction would make it a nightmare.


Now for some embarrassing retractions and/or adjustments.

Hopefully at the end of a long and boring post like this no-one will notice.

Quote:

j-curve: I did an experiment with the simulated graphic today, dialling up 6 6 6 on three sliders... Ignoring the side bands, the response came out at 9 10 9 as predicted by the 1-2-6-2-1 effect... even the most aggressively curve-fitting manufacturer won't achieve 0-1-6-1-0...

Using a graphic (as opposed to parametric) EQ simulator without bandwidth correction the result comes out at 7 7.5 7, so 0-1-6-1-0 is effectively achieved. There is however 3dB of spillover into the sidebands, so 1-2-6-2-1 is not entirely dead. As you might expect, increasing the boost makes things a little worse. For example:-
0 0 0 12 12 12 0 0 0 gets you
1 3 8 14 14 14 8 3 1.

Quote:

Joe Bloggs: ...if you have an EQ with all bands set to -6dB..., it would just come out as -6dB on all bands, right? If you go calculating the effect on adjacent sliders by the absolute amount of cut, you'd conclude that you get -9dB on all bands, which just doesn't seem right... j-curve: Nope, it's absolutes, and it would be -12dB, not -9dB.

Wrong again, *Doh!* Sorry about that. Absolutes are for series cascaded filters.
If it's a parallel bandpass summing graphic EQ (with no bandwidth correction) it comes out as a wobble between -7 and -7.8 dB.

==========

Finally for some entertainment, some interesting articles by Dennis Bohn, EQ guru at Rane. Exposing Equalizer Mythology deals with common misconceptions about EQ and its effects. Do people agree with what he is saying in this article?
And if anyone is still awake, then this great but somewhat technical article on constant-Q equalizers will certainly help.

 

[Joe Bloggs]

Well, I still don't know what I'd do if faced with a 'constant Q' equalizer, but now I know for sure that whatever I need to do there, it has nothing to do with the Shibata EQ:

http://www.audio-illumination.org/fo...=ST&f=1&t=3531

I don't have to guess what the actual settings are. The attenuation / boost at each centre frequency is exactly represented by the sliders.

 

[j-curve]

Hmmm, KikeG doesn't say how he reached this conclusion or whether he did any tests. What happened to the Joe Bloggs who steadfastly held his line in this thread until I became educated and started making sense? Are you going to take this guy's word without any further explanation? It's one thing to code up a digital filter from a textbook but completely another thing to understand how the monster works and what response plot it generates. [BTW I don't claim to have done either yet].

Being the antagonist I should do the test and post the results but I'm faced with various obstacles:-
- the link to Shibata-san's EQ seems to be dead.
- I don't use WinAmp and it's a huge download in which I have no interest other than to investigate the EQ(!)

Here's what I would do if I had the setup:-
I would record some white noise with flat EQ, then take 3 sliders, set them to -12 0 -12 somewhere in the middle of the band and record a sample of this filtered white noise. Then I'd play back the white noise into Spectrogram and use the frequency calibration function until it measures flat (to make up for the soundcard and whatever else adds its own special flavour on the way). Finally I would play back the filtered noise into (freshly calibrated) Spectrogram and check the response at the three centres X Y Z using a log plot and a long Spectrum Average of say 100, so the curve doesn't bounce around too much.

If the XYZ plot and its surroundings came out something like:-
-2.3 -6.5 -12.3 -6.4 -12.3 -6.5 -2.3
then I would know it's a standard graphic EQ with no bandwidth correction.

On the other hand:-
-0.7 -3.0 -12.1 -2.4 -12.1 -3.0 -0.7
would imply 50% bandwidth correction (ie: filter widths only half of the band spacings).

To get rid of the -3.0 and -2.4 spillover as KikeG is proposing, you would need even higher Q, which would likely result in ringing. This seems entirely possible since I think you mentioned ringing before.

 

[j-curve]

Forgot to mention, if you'd like a copy of the spreadsheet I put together just PM me with your email address. That goes for anyone else too.
The spreadsheet is no work of art but it gets the numbers (I hope!) for regular graphic EQ's and chained parametric EQ's.