Calibration tool for Headphone + Hearing capability

[Peter Verbeek]

There are cases where you benefit by dynamic adjustment - e.g. head tracking for spatializing algorithms - and this can meaningfully improve subjective experience, but for frequency response correction, a static equalization may be applied.

So that isn't a problem as it's static data.

Specifically, my recommendation would be using a diffuse field equal loudness - either a directly sourced one from the lit (there's one somewhere in the ISO standards, I believe, maybe even the same work as the FF equal loudness), or by applying the delta of FF > DF HRTF to the FF equal loudness (although this may have issues, due to the aforementioned influence of perceived acoustic source position) - and, following the work of Olive, Welti, McMullin, and Khansaripour, applying a generally "downward sloping" adjustment.

In test interface I've used the equal loudness contour ISO 226:2006. A user can select a Phon (0 to 100) to compare his test results against. From some experimenting the 60 Phon seemed to produce the best eq as the 0 Phon, the hearing treshold, producted too much bass in the resulting created eq. As I don't have any knowledge in this field I don't understand this. A better equal loudness contour could certainly help. Somehow the best results of the equal loudness contour is in the middle section of our hearing. Perhaps the equal loudness contour doesn't take headphones into account as shown in the image of figure 17. Then again I don't know how the measurements were done for creating the equal loudness contour of IS0 226:2006.

Anyway, if I offer all kinds of different equal loudness contours users might be lost. Somehow we need to come up with only a few which users can use. For advanced users there can be more like in different types. Also as this isn't my expertise I've a hard understanding it all. It's difficult to extract a test interface from all these different ways at looking at the research of the human hearing and their results such as the equal loudness contour. I'm sorry for my lack in knowledge and understanding.

You could allow listeners to adjust a "slope" parameter (e.g. the stylized form of the original Harman adjustments from 2009 are 1dB/octave from 20-20000hz), give listeners the obligatory two 2nd order shelf filters at 105 and 2500hz from the linked paper, or directly apply the average preferred adjustments from said filters that Olive, Welti, and McMullin found:

That would be the explanation why the 0 Phon lacks result into an eq lacking enough bass. I guess doing a test with speakers in ones room would result into an eq (0 Phon based) with more bass. I'm speculating of course. Yes, such curves as shown in figure 16 can be used by the user to introduce more more/less bass and/or treble in the resulting eq.

so I would encourage you to include a note suggesting that users adjust the result

In the current version I've already included notes for adjusting the resulting eq (on the interface and in the manual). You're entirely right that this is needed. Perhaps I'll take another look at these notes.

 

[Peter Verbeek]

Under "Accessibility" in the settings, one can navigate to "Audio/Visual" then to "Headphone Accomodations" and then to "Custom Audio Setup". At that point one can select from several presets, but more importantly, one can choose to use a personal audiogram as a basis.

I"ve used the app "Mimi" to generate an audiogram of my aging hearing and confirmed that I have what is defined as a "Slight" hearing loss. Mimi can then save the audiogram to the Apple "Health" app. This audiogram then serves as the basis for a custom EQ for my Airpods Pros.

On Android there are also apps like Neutralizer. This is why we built the test interface as on Windows there was nothing we could find to be very useful.

I'm now tempted to use the Mimi generated audiogram as a basis for an EQ under APO/Peace

An audiogram can easily be used. Just start with a flat eq such as "Equalizer Default", enter the frequencies of the audiogram in the fields above the sliders and do the reverse as the audiogram is stating: There where's a dip in the frequency on the audiogram dial up the slider. After doing this for all frequencies you might need to dial down the pre amp a bit (top slider) to compensate for the sliders you have dialed up.

 

[mmdiss]

...while I am looking for clues on how my headphones should sound (I'm not using Peace to correct a hearing deficit, I'm just trying to squeeze everything out of my inexpensive headphones.... and you can understand is challenging if you don't know in the first place how a proper headphone should sound), I realized that, even if I don't have any of the "supported" devices listed in the AutoEQ project, I might have similar models (for example, the Plantronics 325T listed in AutoEQ looks very, very similar to my Plantronics 3220; also, I own a pair of KZ ZS4, which are clearly different from KZ ZS3, the latter included in the AutoEQ listing, but I managed to compensate the ZS4 frequency response to the ZS3, from graphs found on the net, only to apply later the AutoEQ filters). How much flat is the output obtained using a predetermined correction for another device that you think is equal (or how accurate is the correction from a device to a similar one, before applying the correction) is anyone's guess...

Needless to say, these results are more confusing than ever for me. It looks like the output of equalizing with the equal loudness curves generates an excess of high frequencies and a lack of 4-8khz. Few posts ago someone said about a peak in 8khz which, when compensated, generates a weird sound...Well, during the last few months, I've got accustomed to that, I even enjoyed that sound for as much I deluded myself that was the sound really intended by the music producers. Now I understand that maybe I was just completely off the track :O

A few hours ago, after some kind of "reverse engineering" of the charts taken from AutoEQ, I realized that together with the Equal Loudness curve we might need to include (sum) one of the Harman curves. Does this make any sense? Is this how the Harman curve should be intended? ("on top" of the Equal Loudness "concept")

 

[Mad_Economist]

In test interface I've used the equal loudness contour ISO 226:2006. A user can select a Phon (0 to 100) to compare his test results against.

[...]

Anyway, if I offer all kinds of different equal loudness contours users might be lost. Somehow we need to come up with only a few which users can use. For advanced users there can be more like in different types. Also as this isn't my expertise I've a hard understanding it all. It's difficult to extract a test interface from all these different ways at looking at the research of the human hearing and their results such as the equal loudness contour. I'm sorry for my lack in knowledge and understanding.

My recommendation would be offering only a single equal loudness contour. Given that you're doing a threshold-of-hearing test, this should be the 0 Phon curve, and the one I would suggest is the diffuse field one from ISO389-7 - this used to be part of 226, which I believe is what I was thinking of.

I have included it in .csv format, along with all the data I have displayed as a .xlsx, attached to this post as a zipped folder.

Note that my copy of ISO389-7 is old (from 2006), so if someone feels inclined to buy the latest revision to check if there's been a slight adjustment to the estimated thresholds, that wouldn't be necessarily ill advised.

I guess doing a test with speakers in ones room would result into an eq (0 Phon based) with more bass. I'm speculating of course.

This is a common misconception - the human tendency to prefer a "downward sloping" response doesn't appear to correspond to anything other than...us liking it, roughly. We like speakers whose response and directivity produce something like it, and we like those speakers more when we equalize them to be more like that. We like headphones that sound like those same speakers.

From some experimenting the 60 Phon seemed to produce the best eq as the 0 Phon, the hearing treshold, producted too much bass in the resulting created eq.

So your finding is roughly that the "threshold of hearing" at low frequencies is too sensitive, if I'm understanding correctly? Does this hold using band-limited white noise centred on the frequency in question instead of sines?

Edit: Wait, I have that backwards, you've got insufficient sensitivity? Are you certain that the results are incorrect? I can measure the results if you can send me a couple of your adjustments.

 

[Robbo99999]

...while I am looking for clues on how my headphones should sound (I'm not using Peace to correct a hearing deficit, I'm just trying to squeeze everything out of my inexpensive headphones.... and you can understand is challenging if you don't know in the first place how a proper headphone should sound), I realized that, even if I don't have any of the "supported" devices listed in the AutoEQ project, I might have similar models (for example, the Plantronics 325T listed in AutoEQ looks very, very similar to my Plantronics 3220; also, I own a pair of KZ ZS4, which are clearly different from KZ ZS3, the latter included in the AutoEQ listing, but I managed to compensate the ZS4 frequency response to the ZS3, from graphs found on the net, only to apply later the AutoEQ filters). How much flat is the output obtained using a predetermined correction for another device that you think is equal (or how accurate is the correction from a device to a similar one, before applying the correction) is anyone's guess...

Needless to say, these results are more confusing than ever for me. It looks like the output of equalizing with the equal loudness curves generates an excess of high frequencies and a lack of 4-8khz. Few posts ago someone said about a peak in 8khz which, when compensated, generates a weird sound...Well, during the last few months, I've got accustomed to that, I even enjoyed that sound for as much I deluded myself that was the sound really intended by the music producers. Now I understand that maybe I was just completely off the track :O

A few hours ago, after some kind of "reverse engineering" of the charts taken from AutoEQ, I realized that together with the Equal Loudness curve we might need to include (sum) one of the Harman curves. Does this make any sense? Is this how the Harman curve should be intended? ("on top" of the Equal Loudness "concept")

You could just buy one of the headphones that Oratory has measured that fits within your budget, then use his Harman EQ. And I'd check other sources of any headphone you look at to find out if there is any longevity or fitment issues, etc.

 

[Peter Verbeek]

A few hours ago, after some kind of "reverse engineering" of the charts taken from AutoEQ, I realized that together with the Equal Loudness curve we might need to include (sum) one of the Harman curves. Does this make any sense? Is this how the Harman curve should be intended? ("on top" of the Equal Loudness "concept")

Thanks for sharing your experience. Looking at the Harman curves it seems to be about compensating those frequencies (low/bass and high/treble) that headphones, or perhaps wearing headphones, are struggling with. That's why the in-ear curve has a higher bass part as there isn't much room for those frequencies in the first place. If the equal loudness contours are determined by using headphones the frequencies of a Harman curve are more or less already there in the contours. But if the equal loudness contours are determined by using speakers then the Harman curves can be "added" as you're implying. Of course this is also about taste of a listener. Also it could be that the equal loudness contours are derived from measurements with speakers and headphones. For instance, determining a equal loudness contour at lower frequencies than 500 Hz is difficult by using speakers. During the Fletcher-Munson reseach (1933) this probably was very difficult, almost impossible with the materials available back then.

I'm just trying to squeeze everything out of my inexpensive headphones....

Yes, that's a challenge. The frequency response might differ between types of headphones such as the Plantronics 3220 and 325T, and the KZ ZS4 and ZS3. I think for "sqeezing everything out" this isn't the best approach when you don't have a frequency response precisely of your brand and type. But then again using the test interface of Peace might not do the trick either. This interface can be used to determine in large where hearing problems lie (headphones and/or ears) but for finetuning it isn't a good tool. The usage of equal loudness contours makes it too general. On the other hand, having the exact frequency response of your headphones (by Oratory for instance) might not be enough too. You still need to adjust for taste and perhaps music genre. Anyway, before the test interface I've advised users of Peace to play their music genre and just dial up a slider whilst listening just to see what that particular frequency is adding to the mix. Needless to say that this is very rough way of doing equalization. For cheap headphones this is working but for getting the most out of expensive ones this isn't a good way of doing things.

 

[Peter Verbeek]

My recommendation would be offering only a single equal loudness contour. Given that you're doing a threshold-of-hearing test, this should be the 0 Phon curve, and the one I would suggest is the diffuse field one from ISO389-7 - this used to be part of 226, which I believe is what I was thinking of.

Thanks for the xlsx and csv.
Looking at the small difference in SPL between the free field and diffuse one this wouldn't compensate the resulting eq in the bass part when using 0 Phon. I'm still left with the question why 0 Phon produces a high bass eq. Of course looking at the Phon graphs, their slopes upwards in the low frequencies is quite remarkable. Could it be that SPL just can't simply be used as a difference (delta) in gain, I mean between what's being measured by the user and an equal-loudness contour. Or is it headphones biased? Or simply that it's correct but we like to have more bass in our headphones (as you're implying I guess, see my remarks below)?

This is a common misconception - the human tendency to prefer a "downward sloping" response doesn't appear to correspond to anything other than...us liking it, roughly. We like speakers whose response and directivity produce something like it, and we like those speakers more when we equalize them to be more like that. We like headphones that sound like those same speakers.

Good point. I guess you're saying we like to have bass?

So your finding is roughly that the "threshold of hearing" at low frequencies is too sensitive, if I'm understanding correctly? Does this hold using band-limited white noise centred on the frequency in question instead of sines?

Without a background noise a user listening if a sine (or any sound for that matter) is present, just needs a very quiet room. If a room isn't sufficient quiet the test result will be useless. But using a background noise whilst listening to a sine compensate this. The background sounds may be louder but lower than the noise. But using background noise for sines might not be ideal too. Or sines in the first place. Unfortunately I don't know how to create narrow-band white noise samples for the frequencies to be tested. Besides, I don't know how narrow the bands should be of these noise sample. Too wide and more than the target frequency is going to be tested, too narrow and it isn't noise to begin with.

Edit: Wait, I have that backwards, you've got insufficient sensitivity? Are you certain that the results are incorrect? I can measure the results if you can send me a couple of your adjustments.

The results might perfectly alright. I haven't thought about that. As you're explaining the resulting eq could be perfect but just lacking bass which a headphones manufacturer puts in into his headphones (by using his kind of Harman curve). Anyway, I have tested my Beyerdynamic DT770 32 Ohm velour cushions. It seems that using 20 Phon for creating an eq is the best setting for me. Of course this is near 0 Phon which could be alright too. The test was done with background noise for the sines. I've attached the test file and the resulting eq.

 

[Thalis]

I am going to try this again for:

AKG K240 MkII with velour pads
Philips SHP9500 stock
Samsung EO-IG955 AKG with memory foam tips

I see Phon 60 is on by default? Or which should I use and will it be different for IEMs?

 

[Peter Verbeek]

I see Phon 60 is on by default?

Yes, 60 Phon is default as this commonly yielded the most neutral bass in an eq created from the test results. In other words, heaving to little or to much bass in the resulting eq is just a matter of selecting a different Phon and create the eq again.

Or which should I use and will it be different for IEMs?

In ear certainly a matter on its own. I would go with 0 Phon to see if the bass is well presented in these IEM's.

 

[Thalis]

Yes, 60 Phon is default as this commonly yielded the most neutral bass in an eq created from the test results. In other words, heaving to little or to much bass in the resulting eq is just a matter of selecting a different Phon and create the eq again.


In ear certainly a matter on its own. I would go with 0 Phon to see if the bass is well presented in these IEM's.

Previously when i used Phon 60, I got practically no bass . The curve looked like this:

Will try Phon 0 for the IEMS later.

 

[Peter Verbeek]

Previously when i used Phon 60, I got practically no bass .

Yes, that's possible. Therefore I recommend, in the manual for instance, to only test the middle part of ones hearing. This is the part that is the most important to correct. Bass for instance is usually a matter taste. Anyway, I suggest you try a few Phon's and/or set the starting frequency to 500 Hz for creating an eq.

 

[Mad_Economist]

The results might perfectly alright. I haven't thought about that. As you're explaining the resulting eq could be perfect but just lacking bass which a headphones manufacturer puts in into his headphones (by using his kind of Harman curve). Anyway, I have tested my Beyerdynamic DT770 32 Ohm velour cushions. It seems that using 20 Phon for creating an eq is the best setting for me. Of course this is near 0 Phon which could be alright too. The test was done with background noise for the sines. I've attached the test file and the resulting eq.

I'm sorry, but can you please attach both the result you're getting with the 0 phon curve, and with whatever threshold level is your "preferred"? It is the delta I am interested in.

FWIW, the response shown in your EQ seems, if a bit bass boosted given the DT770's frequency response, not ludicrously out of line in terms of potentially preferred bass response.

Unfortunately I don't know how to create narrow-band white noise samples for the frequencies to be tested. Besides, I don't know how narrow the bands should be of these noise sample. Too wide and more than the target frequency is going to be tested, too narrow and it isn't noise to begin with.

You want a bandpass centered on the band-centre frequency (the one you're doing as a sine at the moment). The order of the bandpass is to some degree preferential - if you had very narrow bands (say 1/12 octave), I would suggest a fairly sharp drop-off in the stopbands, whereas with a 1 point per octave measure, you are already obviously incorporating a lot of "leak".

Could it be that SPL just can't simply be used as a difference (delta) in gain, I mean between what's being measured by the user and an equal-loudness contour.

Simply, yes. Equal loudness is impacted by both level - as shown in the ISO 226 curves - and "circumstance" of the sound field (as shown in the ISO 389-7 curves). One could attempt a very coarse inference of HRTF+hearing effects with two listeners in the same sound field, in a room with exceedingly low noise, and with the stimulus at the same level. Changing any of those variables makes things more complicated.

Or is it headphones biased? Or simply that it's correct but we like to have more bass in our headphones (as you're implying I guess, see my remarks below)?
[...]
Good point. I guess you're saying we like to have bass?

To be clear, we like more bass in general - speakers are not preferred when the power response is flat, and when in-room speaker power response has the same downward slope we like in headphones, we like the speakers. So, yes, you have it right: Humans, it would seem, are at least somewhat "all about that bass, about that bass".

Without a background noise a user listening if a sine (or any sound for that matter) is present, just needs a very quiet room. If a room isn't sufficient quiet the test result will be useless. But using a background noise whilst listening to a sine compensate this. The background sounds may be louder but lower than the noise.

As mentioned above, the real level of the test sound matters for the relative response, so it's quite important that listeners use the actual threshold of audibility, 0 Phon, rather than masking the sound with additional noise and raising the volume. This may require finding a quiet room, but average room acoustic noise should not be above the masking threshold at most frequencies.

 

[mmdiss]

If the equal loudness contours are determined by using headphones the frequencies of a Harman curve are more or less already there in the contours. But if the equal loudness contours are determined by using speakers then the Harman curves can be "added" as you're implying.

Thanks for the reply! It appears that the original equal loudness curve was estimated, back in 1933, using headphones (taken from Wikipedia).

However, the current set of equal loudness curves by ISO 226:2003 seems to be based on "free-field listening conditions" (which translates to "loudspeakers", am I right?). Taken from here https://copublications.greenfacts.o...l-music-player-mp3/figtableboxes/figure-1.htm

As of today, we basically don't have a "real" equal loudness curve for headphones. It is like saying that a flat tuned headphone (that is, each tone being perceived equal considering the loudspeaker equal-loudness contour... Like I'm trying to do with Peace), doesn't sound like a flat loudspeaker and, to make that really but wrongly "flat" headphone to sound like a proper flat loudspeaker we need the Harman curve, isn't it? It's like the Harman curve is equalizing for the amplification/attenuation of specific frequencies done by the shape of our ears (which applies if you listen to a loudspeaker but not using a headphone).

Please, anyone, correct me if I'm saying something wrong, I'm really just trying to understand. If I am correct, I'll be persuaded to add the Harman curve to the equal loudness.

 

[Mad_Economist]

However, the current set of equal loudness curves by ISO 226:2003 seems to be based on "free-field listening conditions" (which translates to "loudspeakers", am I right?).

Free field requires a loudspeaker or other non-headphone sound source, but it's not reflective of real loudspeaker behavior. In a free field, only direct sound is present, so you get the axial frequency response of a loudspeaker as the the net system response. In a real room, sound radiates in multiple directions and that energy bounces around (or, for large wavelengths, "wiggles around in") the room. As a result, speaker power response differs from the axial frequency response.

As of today, we basically don't have a "real" equal loudness curve for headphones. It is like saying that a flat tuned headphone (that is, each tone being perceived equal considering the loudspeaker equal-loudness contour... Like I'm trying to do with Peace), doesn't sound like a flat loudspeaker and, to make that really but wrongly "flat" headphone to sound like a proper flat loudspeaker we need the Harman curve, isn't it? .

There actually is a (threshold of hearing) equal loudness contour for (a specific) headphone: ISO 389-8 has data for the Sennheiser HDA200. Vencovsky et al 2018 contains the same data for a few audiometric headphones, and the HD650.

Equal loudness is not a useful measure for correction of playback response, however - not in and of itself. There are two potential uses to equal loudness comparisons - one, referenced by @Robbo99999 upthread, is Griesinger's methodology of subjective loudness comparison between headphones and speakers, to attempt to infer the delta of listener head-related transfer function and headphone transfer function, which is essentially the "error curve" of headphone response. The other, which seems to be @Peter Verbeek and @Silvian's main interest for this thread, is accounting for hearing loss by comparing a listener's equal loudness contour to a "healthy average individual".

Neither of these will result in a sound that is "like a flat loudspeaker", of course - because an axially flat loudspeaker's response is different in-room. If perfectly performed, the OP methodology in this thread would result in a sound like listening to a flat loudspeaker in an anechoic chamber, with healthy ears, although as I've referenced previously, it's not going to do that well due to issues with how human perception of loudness interfaces with our perception of the acoustic source.

t's like the Harman curve is equalizing for the amplification/attenuation of specific frequencies done by the shape of our ears (which applies if you listen to a loudspeaker but not using a headphone).

The Harman curve's main feature is a generally downward-sloping response, and it is present in the targets for both headphones and loudspeakers, which are very similar.

The difference is the equipment typically used for the measurement: speakers are typically measured with an axially-flat, omnidirectional microphone capsule. Because headphone systems interact with the pinna and ear as a whole, it is necessary to measure them on anthropomorphic measurement systems, which produce the ear gain I believe you're referencing. Measuring a speaker with the same system will still show such a rise, however, and indeed that's exactly what the Harman research was based on. Consider these plots of a flat in-room, at the listening position response from a Revel speaker pair measured on an omni mic

and on a GRAS 45CA, like Amir uses for headphone measurements.

This is the baseline to which the aforementioned downward sloping adjustment is applied.

 

[Robbo99999]

@Mad_Economist , just a quick aside question to your post above, given the following graph you showed, ( ) as being a flat in room response, is there much merit in an individual using that Target Curve as a basis for EQ experimentation and then just applying an adjustable linear downward tilt to that until tonality sounds right? For instance I approximated quite accurately a Linear Tilt Tone Control Filter (linear from 20Hz to 20000Hz) by adding three Q0.5 High Shelf Filters at 63Hz / 632Hz / 6324Hz (which I discovered on miniDSP forums).......so I'm playing with the idea of experimenting with the idea of just applying Tilt Tone Control to that. Or do you think it would be wise to add say a +3dB Low Shelf Boost at 105Hz to that as well to help compensate for lack of tactile bass, and then experiment with the Tilt Tone Control on top of that? (As an alternative to using Harman Curve as it currently stands.)

 

[Mad_Economist]

@Mad_Economist , just a quick aside question to your post above, given the following graph you showed, ( ) as being a flat in room response, is there much merit in an individual using that Target Curve as a basis for EQ experimentation and then just applying an adjustable linear downward tilt to that until tonality sounds right?

As much so as any other "generally acceptable" baseline, I suppose - if you want that response, something like Oratory's Optimum Hifi target would be similar, as is the classic (and higher resolution) diffuse field.

Or do you think it would be wise to add say a +3dB Low Shelf Boost at 105Hz to that as well to help compensate for lack of tactile bass, and then experiment with the Tilt Tone Control on top of that? (As an alternative to using Harman Curve as it currently stands.)

Subjectively, I think that there is some case to be made for the "leanness" you achieve when the response has a relatively flat band between 100 and 2000hz, vs. a continuous 1dB/oct or similar slope. However, simultaneously, the "smoothness" of a more gradual transition may be preferred. I generally do not apply the Harman bass shelves to my headphones - and indeed do not equalize to the Harman target, preferring the diffuse field target - but if you, unlike me, fall into the largest, most normal of Olive, Welti, & Khonsaripour's three segments, which favours the Harman rise, I'd suggest trying it both ways and seeing which you find subjectively preferable.

 

[Peter Verbeek]

You want a bandpass centered on the band-centre frequency (the one you're doing as a sine at the moment). The order of the bandpass is to some degree preferential - if you had very narrow bands (say 1/12 octave), I would suggest a fairly sharp drop-off in the stopbands, whereas with a 1 point per octave measure, you are already obviously incorporating a lot of "leak".

That's good info. It seems a workable option. However, I also meant that I don't know how to generate this type of noise programmatically and also within a narrow band. And then there's the speed issue as AutoIt is so terrible slow compared to C or C++. With some clever programming I managed to generate the needed sines just fast enough.

I'm sorry, but can you please attach both the result you're getting with the 0 phon curve, and with whatever threshold level is your "preferred"? It is the delta I am interested in.

I'm not sure what you're asking but just to be clear what a test is, I'll explain what a test file .peacetest contains. For each test frequency there's a volume in dB. This paint a picture over a large frequency band, say 200 Hz to 20 kHz as these volumes will differ per frequency. So these values are "measurements" of ones ear or ears that has no correlation yet with the equal-loudness contour. This measurement is drawn as a graph on the test interface so a user might compared to the selected equal-loudness contour. Only when a user wants to, he creates an EQ from his test results. The current selected Phon contour is then taken as base for the calculation of the eq.
In my test case with the DT 770 I've used a background noise but I could do it without this noise to end up with a measurement of my threshold of hearing.

FWIW, the response shown in your EQ seems, if a bit bass boosted given the DT770's frequency response, not ludicrously out of line in terms of potentially preferred bass response.

Yes, that's right. But looking at the difference in the slopes of the bass of the equal-loudness contours there can be huge difference in the resulting bass of an eq. I still don't fully understand these major slopes or better said how to deal with them. Of course we humans aren't vary capable to hear low frequencies at low volumes hence the slopes getter steeper.

So, yes, you have it right: Humans, it would seem, are at least somewhat "all about that bass, about that bass".

So true. All our talks let me (again) believe that testing the bass part of our hearing isn't particularly useful. It seems to be mostly taste-based.

As mentioned above, the real level of the test sound matters for the relative response, so it's quite important that listeners use the actual threshold of audibility, 0 Phon, rather than masking the sound with additional noise and raising the volume. This may require finding a quiet room, but average room acoustic noise should not be above the masking threshold at most frequencies.

I get that. However, it demands users to do a test in a quiet room. The issue here is that people tend to not take this seriously enough ending up with a less qualitative test. Hence the noise raising the volume level. Of course this also has implications on the test result, that's why I say on the interface that a moderate noise level is preferred. Perhaps I should dial this down by saying a low level is preferred.

 

[Mad_Economist]

That's good info. It seems a workable option. However, I also meant that I don't know how to generate this type of noise programmatically and also within a narrow band. And then there's the speed issue as AutoIt is so terrible slow compared to C or C++. With some clever programming I managed to generate the needed sines just fast enough.

Does it need to be generated? Noise files could be pre-made and saved, for example.

If it needs to be generated, can you use EQAPO's EQ functionality to simply apply a bandpass and generate white noise normally?

I'm not sure what you're asking but just to be clear what a test is, I'll explain what a test file .peacetest contains. For each test frequency there's a volume in dB. This paint a picture over a large frequency band, say 200 Hz to 20 kHz as these volumes will differ per frequency. So these values are "measurements" of ones ear or ears that has no correlation yet with the equal-loudness contour. This measurement is drawn as a graph on the test interface so a user might compared to the selected equal-loudness contour. Only when a user wants to, he creates an EQ from his test results. The current selected Phon contour is then taken as base for the calculation of the eq.
In my test case with the DT 770 I've used a background noise but I could do it without this noise to end up with a measurement of my threshold of hearing.

So I've just measured the outputs from the test with the 0 phon contour matched as best I can (ignoring actual audibility), and I'm now specifically very confused about this specific line:

From some experimenting the 60 Phon seemed to produce the best eq as the 0 Phon, the hearing treshold, producted too much bass in the resulting created eq.

I'm trying to figure out how using the 60 phon curve as the target, which calls for a lower threshold of audibility for bass, can possibly have produced higher bass than the 0 phon in the resulting EQ?

I get that. However, it demands users to do a test in a quiet room. The issue here is that people tend to not take this seriously enough ending up with a less qualitative test. Hence the noise raising the volume level. Of course this also has implications on the test result, that's why I say on the interface that a moderate noise level is preferred. Perhaps I should dial this down by saying a low level is preferred.

The problem is that without the reference point of the threshold of hearing - sans masking - you have no idea what equal loudness contour you should be approximating. In the absence of ability to measure the actual SPL, you need to have people use their real threshold of hearing for this test, otherwise their results will not be accurate even within this limited paradigm.

 

[mmdiss]

I'm trying to figure out how using the 60 phon curve as the target, which calls for a lower threshold of audibility for bass, can possibly have produced higher bass than the 0 phon in the resulting EQ?

I found the same effect claimed by Peter: as your audibility threshold is fixed (ie. it doesn't vary by the curve you choose, let's say it's -20dB at a specific bass frequency) you get more amplification starting from 60 phon (let's say -40dB, which is a gain of +20dB) than starting from the 0 phon (that could be at -30dB, with a gain limited to +10dB).

 

[Peter Verbeek]

Noise files could be pre-made and saved, for example.

No, a user must be able to choose any frequencies he wants to test. As such the test interface needs to generate sines or narrow band white noise.

If it needs to be generated, can you use EQAPO's EQ functionality to simply apply a bandpass and generate white noise normally?

First of all this has to be silence generations. Second it's a difficult way of creating sounds. I need a white noise algorithm which includes narrow band frequency parameters. I must say I haven't yet investigated this thoroughly. I can't imagine there isn't such algorithm on the internet.

I'm trying to figure out how using the 60 phon curve as the target, which calls for a lower threshold of audibility for bass, can possibly have produced higher bass than the 0 phon in the resulting EQ?

Perhaps my sentences are faulty. So I have another go at it.
When using the 0 Phon contour the resulting eq has more bass compared to 60 Phon. In general the bass produced with 0 Phon is too much.
Edit: Ignore the above sentence. Sorry, I've said this wrong again. It's indeed the other way around: 0 Phon produces too little bass, therefore 60 Phon is the default setting as it usually produces a flat response in bass.

The problem is that without the reference point of the threshold of hearing - sans masking - you have no idea what equal loudness contour you should be approximating. In the absence of ability to measure the actual SPL, you need to have people use their real threshold of hearing for this test, otherwise their results will not be accurate even within this limited paradigm.

Yes, you're right. But a test on this Peace test interface is done by amateurs compared with a test done by an audiologist. Being accurate is less of a concern. Of course the threshold of hearing is the ideal but after experimenting with testing with some background noise the results can reveal issues in ones hearing or even ones headphones. If an eq should be created from the test results with background noise, is another matter. We, Silvian and I have discussed to offer or not to offer an interface to create an eq from test results. But knowing my users I'm sure they want to be able to create an eq from their test results. I do strongly advise to use these created eq's as a starting point as they can't be very accurate eq's. Although the hearing impaired might benefit from such eq's. Besides, we already have users who are very pleased with the created eq's from their test results.