• WANTED: Happy members who like to discuss audio and other topics related to our interest. Desire to learn and share knowledge of science required. There are many reviews of audio hardware and expert members to help answer your questions. Click here to have your audio equipment measured for free!

Calibration tool for Headphone + Hearing capability

Silvian

Member
Joined
Oct 20, 2020
Messages
9
Likes
23
Hello everybody,

I've been dreaming for such a tool for many years and now it is part of the Peace EQ interface.
Calibration tool.JPG

Has anyone tried it yet? If you use a Windows PC as a music source, give it a go.
I was part of the development and so I used it the most. If you need more info than what's on my write up, ask:
https://www.beautifulaudio.biz/peace-equalizer-calibration-tool

Cheers
 

Dreyfus

Active Member
Joined
Feb 7, 2019
Messages
247
Likes
296
Location
Germany
Thank you for sharing!
Loudness equalizations get way to little love in the community. I hope that will change with your implementation in Peace. :)

May I ask what the corrected curve does apply? Is it a rough compensation for higher listening levels, resulting in less treble and bass emphasis compared to the "boosted" hearing threshold curve? Comparing the graphs it looks like you added the 100 Phon curve. Wouldn't it be better to choose the average of 80 Phon instead?

I had good success with David Griesinger's loudness equalization procedure in the past, tuning my headphones to my average listening level (skipping the HRTF part). He recommends the use of narrow band noise instead of pure sine tones. To me that makes perfect sense since some headphones come with a very wobbly response which can introduce sensible errors unless you significantly increase the number of bands. This gets particularly critical the higher you move up in frequency where all sorts of narrow band intefereneces appear (depending on the individual anatomy in correlation with the characteristics of the headphone).

Personally I do all my loudness equalizations with narrow band pink noise and 31 bands on my average listening level. Since the lower region is somewhat fiddly with random noise I compare my findings to some steps of sine waves to make sure the bass is correct.

Do you plan to expand the features of the module in the future?
I would welcome a customizable number of bands, a selection of noise vs sine and varying levels for the compensation curve.

Best wishes - keep up the great work!
P48
 

Peter Verbeek

Member
Joined
Nov 6, 2020
Messages
72
Likes
48
Location
The Netherlands
@P48 Reading your response on this topic I see you have some questions and remarks. As this is my first post on the AudioScienceReview forum I hope to give some clarification which are conform the etiquettes of this forum.

First of all this new Headphones and hearing test interface in Peace is version 1.0. Any response/feedback is well-appreciated as it could potentially improve this test interface. So thanks, P48.

May I ask what the corrected curve does apply?
The curve used is the Equal-loudness contour at 0 Phon which is the hearing threshold. This is chosen because of the way a test profile of one's headphones/hearing is built up. By listening to a sine whilst lowering its volume till it becomes unhearable the threshold of a test frequency is determined. This threshold is the same as the 0 Phon contour.
As we found that using the Equal-loudness contour resulted into a very flat bass we decided to introduce more bass by correcting the contour. Of course this is abitrary but there are several reasons for this. Perhaps the most important one is that people searching for "the best equalization" for their headphones want a rather "quick fix". Using an uncorrected Equal-loudness contour is fine but it would mean that a Peace user needs to introduce bass and treble himself. Btw. this is still possible if a user just want that way to go. In a future version we want to offer more types of correction. You can think of a few like "Give me more bass". One note: the Equal-loudness contour itself is arbitray. There doesn't exist something like a common hearing/headphones. Each individual has a unique hearing and every headphones set has its own audio profile.

He recommends the use of narrow band noise instead of pure sine tones
Using noise instead of a sine is one feature which is on the wishlist. Up till today I didn't had any understanding of the difference between using a sine and noise. I guess I'm saying thanks for the feedback :) Any addition info is appreciated. The current test interface has the abbility to play noise whilst testing the sine waves over the chosen frequencies.

Do you plan to expand the features of the module in the future?
I would welcome a customizable number of bands, a selection of noise vs sine and varying levels for the compensation curve.
I guess that's a yes :)
To create a test profile (and equalization from it) one chooses a EQ configuration, either the special made one or another like the 31 bands one. It's up to the user which frequencies to test. The special made one holds more sensitive frequencies in the middle of our hearing besides other frequencies like lower bass.
Yes, more compensation curves, noise vs sine (maybe more kind of noises?) and other features.

Peter
 

Dreyfus

Active Member
Joined
Feb 7, 2019
Messages
247
Likes
296
Location
Germany
Welcome to the community, great to see you joining the conversation! :)

Can you explain how the threshold curve the user adjusts is actually processed by your application?

I did a few tests yesterday and found that the curves I created were very bright, as if they followed an upward slope. To my understanding a hearing threshold curve should have a significant bass boost due to the low hearing sensitivity at minimum level. Here you can see both the 0 Phon and 80 Phon curve (which I pull up as an assumption for an average listening level) matched at 1 kHz:

contour.png


As you can see the 0 Phon curve should lead to a strong bass boost of up to 37 dB which is even further emphasized by the attenuation between 1 kHz and 6 kHz, all compared to the average of 80 Phon. In theory you would have to apply the grey curve to convert the 0 Phon response you detected with the calibration tool to get something that sounds more or less balanced at average listening levels. So you would actually take bass away instead of adding it.

I confirmed the observation above by listening to the sine waves generated by your calibration tool but adjusting the faders in the EQ APO editor instead. Before you ask: I used the graphical equalizer and matched the bands to those of the reference, of course.
The result is pretty much what you would expect looking at the graphs above: a very boomy profile with recessed upper mids and presence.

At that point I ask myself why the curves I created with your calibration tool are so bright, even with the correction applied.
Speaking of correction, here is the compensation curve of your calibration tool (1.0) compared to my personal preference:

Corrections.png


I may add a little bit of treble attenuation later. Needs more data for verification.

And some eq curves I generated for my Audioquest Nighthawk (Brainwavz Hybrid):

uncompensated.png


compensated.png


There were a few hours between the two tests. So there may be some discrepancies due to the headphones seating and physiological fatigue. I may repeat the test in near future to gather more data and get more confident with the technical procedure.
However, what you can see already is that my personal preference curve (which is the smoothed average of several loudness experiments I did in the past weeks) is right in middle between the first and second run with your calibration tool for 300 Hz to 10 kHz. The bass is lacking, as already mentioned. The upper end around 10 kHz is really hard to judge because of its high pitch. I boosted those frequencies until I was really sure I could hear the sizzle turning the tone on and off. As it turned out the gain for the air frequencies is much higher than I prefer in practice, listening to music. I may include that into my correction in case the phenomenon keeps to persist.

I think the procedure could definitely profit from narrow band white or pink noise (white = neutral curve, pink = brighter curve) as a test signal. As with the room modes for loudspeakers it makes more sense to excite a broader area of the spectrum to counterbalance dips and peaks in the spectrum. This is especially important for the higher frequencies where the headphone and the individual's ear morphology unpredictably interact with each other. Such effects could easily be skipped with a lower number of bands and a sinusoidal test tone.

Overall the calibration tool has a lot of potential. I am excited about the further development of this project and really appreciate your openness to the community! :)

Best wishes,
P48
 

Peter Verbeek

Member
Joined
Nov 6, 2020
Messages
72
Likes
48
Location
The Netherlands
Can you explain how the threshold curve the user adjusts is actually processed by your application?
As the Equal-loudness contour at 0 Phon resulted in a low bass and a bit low treble I had to introduce a corrected contour. But this corrected contour seems to look like a contour in the range of 40 to 80 as your graph and tests are showing also. Perhaps my initial thought of taking the 0 Phon contour wasn't a correct one although logical as the test process is basically along the hearing threshold. Btw. thanks for thoroughly testing. It really gives the feedback we were hoping for. We need peope to improve what we have come up with. Now for the processing of a test profile to an equalization. This is also rather difficult. It involves these steps:
  1. Getting the differences between the curves of the test ears and the corrected or uncorrected curve (depending what the user has chosen).
  2. Calculating the averages of these differences in order to get a baseline, say a 0 dB point, preferely somewhere in the mid frequencies.
  3. Interpoling the non-tested frequencies. The user might have skipped a few frequencies without testing them.
  4. Creating an equalization based upon the user settings on the "Create equalization window" for the both ears or left and right ear depending on the testing of both ears together or ears separately.
By default the GraphicEQ of Equalizer APO is selected so an equalization is more precisely following the calculated equalization curve. Because by using peak filters as an equalization method the curve will defiate as the gains of the peak filters will be added to each other, especially at lower quality values (broader bandwidths). Ideally for peak filters equalization some intelligent calculation is needed but I haven't yet been able to come up with such a calculation which needs to calculate a frequency based on all other already calculated frequencies which aren't calculated yet. It's like looking into the future for doing a calculation in the past. Anyway, that's for a next version perhaps.

narrow band white or pink noise (white = neutral curve, pink = brighter curve) as a test signal
As I'm not an expert on noise, especially on creating a proper noise signal of any type, I've used a very basic way of creating noise: random samples. It does produce noise but I don't even know what kind. And there's also the issue of using AutoIt, the programming language Peace is written in. It isn't very fast (interpreter language) so I had to deal with its slow processing power. Besides others, this resulted into some issues at the very high frequencies roughly above 12 kHz. The sines aren't as crisp as they should be. Fortunately, it's at very high frequencies but still it may have some negative influence on the test profile. Therefore Silvian and I are thinking of an improved version using another programming language. First choice is c# which I hope is fast enough but it seems it isn't that easy to generate sines and noise either. So it's a work in progress.
Anyway, do you know how to generate narrow band white and pink noise? And I'm wondering, I've chosen a sine for its specific frequency. Noise is all over the place making it difficult to get a test profile for a particular frequency. Any thoughts on this matter?
 

solderdude

Grand Contributor
Joined
Jul 21, 2018
Messages
15,891
Likes
35,912
Location
The Neitherlands
Sorry, ik weet wat een wobble is. Is het, in dit geval, ruis welke in volume verandert in de tijd?

I'll reply in English so others can better understand what we are talking about. :)

A tone shifting in frequency pretty rapidly (you can even use noise to modulate.
With a wobble (warble) you can sharply define the band in which the test tone is.
Noise bands require steep slopes, wobble tone is sharper defined.
Basically a VCO with a small range.
Technics had a consumer analyzer that used this, the SH8000
It was basically a tone generator with a warble (to create bands) and an SPL meter.
A speaker calibration tool that didn't need a display and was small.

Just like noise bands a wobble/warble excites all frequencies in a narrow frequency band so even when there is a sharp null or peak it will be evened out in 'power' leading to a more averaged measurement without the need of sharp filtering noise.
 
Last edited:
OP
S

Silvian

Member
Joined
Oct 20, 2020
Messages
9
Likes
23
As the other participant in developing this calibration tool, besides Peter, the brain behind it, I feel I should perhaps explain where I was coming from, as a not so much a technical (how to) person, but as someone who wanted such a tool for a very long time to simply enjoy my listening better. Some of these things are in the write up I mentioned in the first post, but I guess there will be a few readers of this thread who will not necessarily follow that link and also who might be too impressed with the technical details already displayed to give this tool a try.

I am old enough to remember the 30 bands equalizers of the past - I never bought one because I always had this nagging question: how would I know how to adjust it? But I also remember the simple Loudness button all the amplifiers had at that time - I would use it when listening at low levels, as it was intended. I also noticed the approach the likes of Schiit (Loki) and RME (ADI2DAC) have taken when implementing equalizers: a handful of (meaningful) bands. The Descriptors graph Solderdude allowed me to use was another hint that for such a tool to be effective, we don't need to go too far into the number of bands. As for which Phon equal-loudness curves to use, the differences between them are mostly at the extremes (bass and treble) and so getting the mids accurate and then having the user play with a few bands for bass and treble to match their preferences (age, tiredness, recordings, etc) resulted, for me, in what my goal was: the most easy yo use and effective equalizer. By "effective" I mean a tool which tells me what is best for me, in a objective way.

Of course I am looking forward to improving the tool by say increasing the number of bands, mostly in the midrange. I would also like to see a loudness "button" of some sort (to make things even easier for that final adjustment). Ideally the loudness will be applied according to the listening level which is impossible since the amplifier will be after the PC - but we have a few ideas on that too.

Again, the point I was trying to make now for those who didn't try it yet was that the tool is, to me, after 35 years of interest in Hi-Fi, the best and easiest way to calibrate the pair of one's hearing and a particular set of headphones.
 

Attachments

  • descriptors2.png
    descriptors2.png
    88.9 KB · Views: 263
  • phon.JPG
    phon.JPG
    51.4 KB · Views: 239

Dreyfus

Active Member
Joined
Feb 7, 2019
Messages
247
Likes
296
Location
Germany
As the Equal-loudness contour at 0 Phon resulted in a low bass and a bit low treble I had to introduce a corrected contour. But this corrected contour seems to look like a contour in the range of 40 to 80 as your graph and tests are showing also. Perhaps my initial thought of taking the 0 Phon contour wasn't a correct one although logical as the test process is basically along the hearing threshold.
[...]
Thanks for the remarks.

To my understanding the correct way would we to subtract the difference between the 0 dB curve and the 80 Phon curve from the raw input data.
Did I get it right that you used the 0 Phon curve straight away as a compensation curve? That would explain the lack of bass for the curves I got out of your calibration tool.

As I'm not an expert on noise, especially on creating a proper noise signal of any type
Me neither, unfortunately.

David Griesinger recommends 1/3 octave noise as referenced in the IEC 60268-7 standard. Maybe someone with access to the papers can look it up?

Regards,
P48
 

Peter Verbeek

Member
Joined
Nov 6, 2020
Messages
72
Likes
48
Location
The Netherlands
Did I get it right that you used the 0 Phon curve straight away as a compensation curve?
Based on the 0 Phon curve I build a correction curve to compensate the lack of bass and treble. This process is arbitrary but it was needed to get the first version of this tool out there. In other words, without it the 0 Phon curve would yield a bassless equalization. And more importantly, the notion that a user would think that he/she can't hear properly in the bass and treble regions. Having correction/compensation curves will point the user to the proper understanding how to use this tool.

I think we have to experiment with the 40 to 80 Phon contours just to find out what equalizations they yield and of course what it looks like in the graph of the test interface.

To my understanding the correct way would we to subtract the difference between the 0 dB curve and the 80 Phon curve from the raw input data.
I'm not entirely sure what you're saying here. Do you mean we should use the difference between the 0 Phon and 80 Phon curves? Or the difference between the test data curve and the 80 Phon curve?
 

Peter Verbeek

Member
Joined
Nov 6, 2020
Messages
72
Likes
48
Location
The Netherlands
Just like noise bands a wobble/warble excites all frequencies in a narrow frequency band
Do you mean playing random frequencies amongst a center frequency? For instance, center frequency (to test) 1 kHz, bandwidth 200 Hz, test sound generated noise in band between 900 and 1100 Hz.
Or do you mean a fixed wobbling around the center frequency? Taking the above figures, a sound which oscillates nicely between 900 and 1100 Hz.

I can see both as useful but that's just intuitively speaking. My lack of knowledge on this subject is evident :)
 

solderdude

Grand Contributor
Joined
Jul 21, 2018
Messages
15,891
Likes
35,912
Location
The Neitherlands
Yes, those 2 are basically the same.
You can frequency modulate (shift) in a sinusoidial way or determined by random noise.
Of course when you sweep over a wobbly part of the frequency range you do get 'AM detection' so one would have to find an optimal 'modulation' and band where the hearing can deal with.
The main thing is (aside from the AM demodulation) that the frequency band is tighter and no overlap from neighboring bands is there.

It's just another method to obtain the same thing. Determine the frequency response by ear minimizing potentially problematic FR issues of drivers/conditions.
 

Peter Verbeek

Member
Joined
Nov 6, 2020
Messages
72
Likes
48
Location
The Netherlands
Yes, those 2 are basically the same.
I understand. It's more about avoiding a singular sine frequency and using a frequency band to improve the overall quality of the test. I don't think I can implement this in the current test interface in Peace. AutoIt, the programming language I use, is to slow for this. This is a nice feature for a next version which needs to be build with a faster programming language.

you do get 'AM detection' so one would have to find an optimal 'modulation' and band where the hearing can deal with.
Sorry, I can't follow what you're talking about. Is this because doing a wobbling in a frequency band the user perceives this as an increase in the gain (like say compression)?
 

solderdude

Grand Contributor
Joined
Jul 21, 2018
Messages
15,891
Likes
35,912
Location
The Neitherlands
Sorry, I can't follow what you're talking about. Is this because doing a wobbling in a frequency band the user perceives this as an increase in the gain (like say compression)?

It's a driver issue. Drivers (speakers as well as headphones) can have sharp dips and peaks. In most measurements smoothing is done so these narrow peaks and dips are 'averaged'.
This is what warble and noisebands do as well.
So the wobble is another form of applying smoothing (or noise bands).

When using a wobble the sine is swept over a band. Due to the drivers response loudness of the sweeping tone varies in amplitude (AM = Amplitude Modulation) and you might also 'hear' the shift in frequency as a separate tone.
So why I suggested a modulation by LF noise instead of a sine (usually it is a sine).

When one were to use a sine at a fixed frequency and you happen to hit a narrow peak or dip in the driver you will perceive it louder or softer.
When it is just one and you want to determine a 'smooth' curve (which would make sense) you can also ignore or interpolate.

But as currently you cannot implement it is just to let you know there is another method compared to sine or narrow band noise.

Should you want to have a closer look at this you can always contact me (in Dutch) via PM.
 

Peter Verbeek

Member
Joined
Nov 6, 2020
Messages
72
Likes
48
Location
The Netherlands
It's a driver issue.
Right. In all the years I've been working with Equalizer APO (helping people and using it for Peace) sometimes someone comes along complaining about some sort of compression done by the driver. Any thoughts if related to what you're saying about drivers here? Is it the loudness correction or response of a driver?

Thanks for the good explanation of the usage of a wobble and the usage of noise. Great feedback. I'll PM you if or rather when I'm able to build the new test interface and/or improve the existing one. Thanks for the offer.
 

solderdude

Grand Contributor
Joined
Jul 21, 2018
Messages
15,891
Likes
35,912
Location
The Neitherlands
When it comes to compression from headphone drivers then there are only a handful that do that and its mild and only in the lower frequencies at high SPL.

It is not related to dips peaks > 300Hz.
 

Dreyfus

Active Member
Joined
Feb 7, 2019
Messages
247
Likes
296
Location
Germany
Based on the 0 Phon curve I build a correction curve to compensate the lack of bass and treble.
As already mentioned, the 0 Phon curve and everything that follows that (such as a loudness test at minimum level) is bass-boosted per se. When doing such tests our hearing needs much more energy in the lower spectrum due to our ears lack of sensitivity in that range. That is what is being described by the loudness contour and leads to a curve pointing up towards the lower end of the scale. When your 0 Phon equalization procedure causes a curve that actually reduces the bass instead of boosting it, then there must be something wrong with your processing.

That's all I can say at that point.

Best wishes,
P48
 

Thalis

Senior Member
Joined
Sep 2, 2020
Messages
362
Likes
217
I tried the test on my K240 Mk2... the curve I came up with... or rather the sound... is ummm... no difference when I compensate with same gain/loudness. I have tried different noise levels and repositioning the cups. So I ended up not EQing the AKGs at all.

Then with the Koss KPH30i... there is a difference. Previously I came up with my own curves trying to match the AKGs and also tried the KSC75 curve. The KSC75 curve was already quite good at opening up the KPH30i. Then I tried the test after reading this post and after many failed attempts ... i think its an improvement... at least to what i am hearing now. I kept having trouble with 4183Hz though. Every test I ended up with a very bright and compressed sound signature and 4183Hz was way up high. So one night when it was extra quiet I cleaned up my ears further and upped the noise level to -40dB and did the test again. Somehow i did not get the same results as previously after repeated tests. Not sure why or how that I got that particular frequency tamed a bit more.

So now the KHP30i sounds great at least to my ears. Very open, detailed, not fatiguing. Not losing any bass detail either. All is good and balanced. I do hope I did the test correctly and used all the frequencies.

At the last stage before saving the curve, I used: 10Hz to 20kHz, EQ Strength=Normal, EQ Type=Graphic EQ
 

Attachments

  • Koss KPH30i-Equalizer APO 05.jpg
    Koss KPH30i-Equalizer APO 05.jpg
    157.8 KB · Views: 293
  • Koss KPH30i-Equalizer APO Curve 05.jpg
    Koss KPH30i-Equalizer APO Curve 05.jpg
    81.9 KB · Views: 271
OP
S

Silvian

Member
Joined
Oct 20, 2020
Messages
9
Likes
23
I tried the test on my K240 Mk2... the curve I came up with... or rather the sound... is ummm... no difference when I compensate with same gain/loudness. I have tried different noise levels and repositioning the cups. So I ended up not EQing the AKGs at all.

Then with the Koss KPH30i... there is a difference. Previously I came up with my own curves trying to match the AKGs and also tried the KSC75 curve. The KSC75 curve was already quite good at opening up the KPH30i. Then I tried the test after reading this post and after many failed attempts ... i think its an improvement... at least to what i am hearing now. I kept having trouble with 4183Hz though. Every test I ended up with a very bright and compressed sound signature and 4183Hz was way up high. So one night when it was extra quiet I cleaned up my ears further and upped the noise level to -40dB and did the test again. Somehow i did not get the same results as previously after repeated tests. Not sure why or how that I got that particular frequency tamed a bit more.

So now the KHP30i sounds great at least to my ears. Very open, detailed, not fatiguing. Not losing any bass detail either. All is good and balanced. I do hope I did the test correctly and used all the frequencies.

At the last stage before saving the curve, I used: 10Hz to 20kHz, EQ Strength=Normal, EQ Type=Graphic EQ

Hi Thalis,

Did you start with the "Headphones and Hearing Test" configuration? The frequencies you have in your photos do not match those in the config I mentioned. Not that there's anything wrong with the frequencies in your test, but there are a few more useful frequencies in that config. Also, in case you didn't read it already, perhaps it is worth looking at how I do the test: https://www.beautifulaudio.biz/how-to-best-use-it. I used a lot of times, on quite a few headphones, and that's the process that works best for me. And yes, ideally it would be quiet enough to actually hear the frequencies, that's a problem I have, lots of traffic very close to my window and sometimes I have problems picking up especially the low frequencies.
If I was to aim for a validation that the tests, for both your headphones, were successful, I would expect, as I said in my write up, that both your headphones to sound a lot more similar than you were previously aware of.
Let us know.

Cheers
 
Top Bottom