Harman preference curve for headphones - am I the only one that doesn't like this curve?

[amirm]

For the half-dozen pairs of headphones I own, I can definitely say that doing my own measurements on a MiniDSP EARS, has resulted in vastly better EQs than using generic measurements/EQs from Oratory/Crinacle et al.

Those guys way over-filter the response given how loose our measurements are.

 

[bluefuzz]

Those guys way over-filter the response given how loose our measurements are.

Yes, and even when making my own simpler filters based on 'generic' measurements, I find filters based on my own measurements to be 'better'.

 

[Wombat]

I don't think the variation between headphones of the same model is significant enough / often enough to be concerned that headphone EQ's based on measurements from people like Oratory are invalid......as in I've had great experiences with massive sound quality improvements by using Oratory's measurements for my own EQ's as well as using his EQ's.....so when Sean was commenting on this topic by saying "all bets are off in regards to what you are listening to" with regards to quality control on sub $300 headphones I find this to be an overdramatic statement on the situation. It doesn't mean you can't get a better experience by using for instance Oratory's measurements, instead it means you might be really quite unlucky and end up with a sample that is significantly different, but I bet you'd have to be pretty darn unlucky for it to be significantly different.....and even if there is some difference then it doesn't mean that Oratory's EQ wouldn't improve it....it probably would because I guess the deviations would be in certain places in the frequency range rather than grossly incorrect accross the whole board. I think he was being overdramatic in his statement and I think some people are taking it out of context.

Get rid of the 'I thinks and don't thinks', 'I've had great experiences', ' I find', 'you might be unlucky'.

Verifiable results please - yours or validated reference work.

 

[Frank Dernie]

Have you ever measured the headphone before and after TRU-NOTE or compare them? Curious what your thoughts are.

PMFJI
I haven't measured them but I did buy some.
It was an "engineering" decision based on my decades long concern that the same headphones probably sound different to different people simply because of different skull and pinna shape and probably the precise position of the cup on the head each time one puts them on.
I hear an improvement when I do the equalisation, that could, of course, be expectation bias.
Wasn't this work an AKG project pre-aquisition by Harman?

 

[Music1969]

THe experimenter said he cannot wear it more than 40 minutes without pain so comfort may be an issue. Too bad because it would be a good headphone for long flights

You should try it yourself next time. He could be an outlier.

No comfort issues for me. Your mileage may vary of course.

 

[MayaTlab]

PMFJI
I haven't measured them but I did buy some.

Tyll Hertsens did some measurements here :
https://www.stereophile.com/content...g-auto-calibrating-over-ear-headphones-page-2

Apple has a number of patents on using various sensors (not mics) inside the earcups to make a more or less rough image of the listener's ears. The most detailed ones use long range capacitance sensors to make a very rough image of the listener's ear and focus on operational qualities. The sensors would have been used to detect orientation of the headphones relative to the user's head (tilt forward / backwards) and left from right ears (and route the proper channel to the appropriate ear). The Airpods Max design was significantly influenced by these patents and remains compatible with them, although obviously none of these sensors made their way into the final product. For example they're the only headphones I've seen in recent memory that are acoustically fully symmetrical front to back (that would have been necessary to make them left / right reversible). Or the fact that most of the earcup's depth is from the solid earcup itself and not the pads, so that they can locate sensors all around the ear (just like what they did with the optical sensor).
In some patents they mention the use of structured light sensors (think Face ID for your ears) to make a higher resolution image of the listener's ear. If the resolution is high enough you can then start to develop customised sound profiles. A number of research teams have done so using various types of algorithms such as neural networks.

 

[restorer-john]

Apple has a number of patents on using various sensors (not mics) inside the earcups to make a more or less rough image of the listener's ears. The most detailed ones use long range capacitance sensors to make a very rough image of the listener's ear and focus on operational qualities. The sensors would have been used to detect orientation of the headphones relative to the user's head (tilt forward / backwards) and left from right ears (and route the proper channel to the appropriate ear). The Airpods Max design was significantly influenced by these patents and remains compatible with them, although obviously none of these sensors made their way into the final product. For example they're the only headphones I've seen in recent memory that are acoustically fully symmetrical front to back (that would have been necessary to make them left / right reversible). Or the fact that most of the earcup's depth is from the solid earcup itself and not the pads, so that they can locate sensors all around the ear (just like what they did with the optical sensor).
In some patents they mention the use of structured light sensors (think Face ID for your ears) to make a higher resolution image of the listener's ear. If the resolution is high enough you can then start to develop customised sound profiles. A number of companies have done so using various types of algorithms such as neural networks.

Obsolete overnight, once we have auditory nerve implants. God help us if "Harman" invents that...

 

[MayaTlab]

Obsolete overnight, once we have auditory nerve implants. God help us if "Harman" invents that...

Well I guess that would be Elon Musk's Neuralink territory ?

 

[bobbooo]

Yes, loudness matching is tricky especially when you have speakers or headphones with widely different frequency responses. It can work for some programs but not well for others. The more homogeneous the program spectra, and the more constant the loudness of the music, the better the loudness normalization will work.


For loudspeaker comparison, we historically used B-weighting (Slow) playing pink noise through the speaker and measuring at the listening seat. In recent years, we adopted ITU-R 1770 loudness. It's not perfect either and there is a question how well it works at different playback levels.

I'm glad you brought up that study where we asked listeners to adjust the level and frequency of the bass shelf filter of the headphone calibrated to the Harman Target. We repeated the measurement where the leakage and the loudness effects from the bass level adjustment were controlled and uncontrolled.

It turned out that the loudness normalization had little effect on the preferred bass level. When leakage was controlled they reduced the level slightly, as you might expect.

There was a small effect from loudness normalization on the bass filter frequency selected.Most people preferred the LF shelving filter to be set below 200 Hz.

But the loudness differences in these experiments are are small in comparison to giving someone a volume control and adjusting the broadband level of music over a range of 60 dB and setting the bass levels. I would expect to see much, much larger effects on where they set the bass level than the effects observed in these experiments.


https://www.itu.int/dms_pubrec/itu-r/rec/bs/R-REC-BS.1770-3-201208-S!!PDF-E.pdf

Thanks for the ITU-R 1770 link. One of the reasons I was asking is I was hoping there might be a fairly straightforward method to calculate the digital pre-amp level for each of several different EQ profiles for a particular headphone to achieve equal perceived relative loudness between them, so a listener could then judge these EQs fairly when blind testing them (at a fixed volume they are generally most comfortable listening at on average with those profiles), in order to remove potential cognitive biases. Loudness normalization between different users whose normal listening level may be very different from each other is another (and seemingly even harder) question, so I was mostly concerned with the first kind of loudness normalization between different EQ profiles with the same user and headphone at a fixed overall volume.

I would not eat at any restaurant with my last name. There is too much negative expectation bias and cognitive dissonance to accept that someone named Olive could cook an edible meal

Now that's commitment to science!

Yes, I am aware of that headphone and the technology. I didn't consider that because it designed more for stereo content-not immersive audio content like ATMOS, MPEG-H etc., which didn't really exist when it was developed.

Have you ever measured the headphone before and after TRU-NOTE or compare them? Curious what your thoughts are.

I don't actually have the N90Q (yet), but I will next week - couldn't resist ordering a pair when they recently went on sale for just $299 (80% discount!), so I can post some impressions then. Apart from Tyll of Innerfidelity's TruNote measurements (the red/orange curves just after 'calibration' on his leg to show the difference, I found the spatial mode measurements more useful and interesting) on his not entirely standard rig already linked to, Head-Fi's Jude also measured them (dotted line after TruNote calibration on an actual person's head, solid on a GRAS KEMAR):

Comparing the bass to (what I believe) are the measurements of the N90Q in your paper, it seems Jude once again didn't get a great seal for this measurement however, although I'm aware this may be down to the KEMAR HATS he used compared to the GRAS 45CA your measurements were made with (and possibly differences in the pinna considering you used custom made ones?). Actually I wanted to ask whether you (or Todd Welti @twelti) remember if you ran the TruNote calibration on the 45CA prior to measuring these headphones? As when I get them I'd like to EQ them to the 2018 Harman target (at least in the bass), so I was hoping to use your measurements of them (if they were made with TruNote enabled) to base the EQ on. I think I remember seeing that they target the 2013 Harman curve, so I suppose I could find out the difference curve between this and the 2018 target and use this as an approximate EQ curve, but I'd prefer to base it on your actual measurements.

Of course you can take measures to minimize those errors by smoothing the measurements, taking averages through multiple reseats which will minimize the Q/gain of the filter applies.

I think this is something those advocating approximated EQ filters based on measurements of a single unit at a single placement are missing. Both their method and that of people like Oratory who average multiple units and placements result in 'smoothed' EQ curves, the difference being the smoothing in the latter is based on averaging actual measured unit and placement variation, whereas the former relies on effectively guesswork for unit variation, and a combination of eyeballing and (fallible) subjective audial judgement of placement variation with all the nuisance variable (e.g. cognitive bias) pitfalls that entails, diminishing the accuracy, reliability and generalisability of EQ profiles created in this way.

I think it should also be considered that a user doesn't necessarily have to use all of Oratory's filters. He usually includes a broad low and high shelf filter to bring the overall tonality to the Harman curve (even stating that the low shelf can be adjusted to taste, which is consistent with Harman's bass preference findings). His other filters are then usually negative-gain peak filters (to avoid inadvertently adding to an unexpected peak, which would be more objectionable then an unintended dip). I think the best approach with his EQ profiles is to apply the shelf filters, then a couple of the main (lowest-Q, highest gain) peak filters, listen, then apply the rest in turn, one after the other, and if at any point the headphones sound worse to the listener (blind listening could even be done), just remove the last applied filter and leave it there. It doesn't have to be an all or nothing affair, and we shouldn't let the 'perfect' (exactly matching the Harman target) be the enemy of the good - bringing the frequency response on average closer to the target and so improving the headphone's sound for the majority of users, units and head placements, via averaging.

 

[Robbo99999]

For the half-dozen pairs of headphones I own, I can definitely say that doing my own measurements on a MiniDSP EARS, has resulted in vastly better EQs than using generic measurements/EQs from Oratory/Crinacle et al.

Is there a valid Target Curve that comes with that kit? How do you use it to create EQ's?

Get rid of the 'I thinks and don't thinks', 'I've had great experiences', ' I find', 'you might be unlucky'.

Verifiable results please - yours or validated reference work.

Get over yourself, I'll speak how I like thanks.

 

[bluefuzz]

Is there a valid Target Curve that comes with that kit? How do you use it to create EQ's?

There are downloadable raw calibrations for the mic capsules as well as a couple of default curves similar (but not identical) to the Harman 2013 and Harman 2018 curves respectively. I prefer to use the raw calibration and load Harman 2018 as 'house curve' in REW. You can of course make your own target curve if you feel so inclined. REW can then export the EQs in various formats you can load (or enter manually) into your EQ software of choice. There is a basic workflow description on the MiniDSP site.

 

[Frank Dernie]

I went to the office today and heard the Max for the first time. I only heard pink noise bursts through it because I was doing a localization test but it was quite impressive. The ANC was very impressive: I could not hear the experimenter giving instructions without taking it off. THe experimenter said he cannot wear it more than 40 minutes without pain so comfort may be an issue. Too bad because it would be a good headphone for long flights

I bought the Max as a matter of interest.
I don't find them uncomfortable but a bit overblown in the bass to give a convincing orchestral balance IMO.

 

[Frank Dernie]

I can't imagine how one could possibly end up with a genuinely "level matched" comparison of headphones with different FR profiles - which is almost all of them as far as I can see.
My scepticism is very high here.

 

[Robbo99999]

There are downloadable raw calibrations for the mic capsules as well as a couple of default curves similar (but not identical) to the Harman 2013 and Harman 2018 curves respectively. I prefer to use the raw calibration and load Harman 2018 as 'house curve' in REW. You can of course make your own target curve if you feel so inclined. REW can then export the EQs in various formats you can load (or enter manually) into your EQ software of choice. There is a basic workflow description on the MiniDSP site.

Thanks, I'll take a look. I've heard there are some inaccuracy issues with miniDSP Ears, but I can't remember what specifically. The method and reasoning behind how they created their Target Curves are also something to be looked into.....I can't comment specifically on how they've done that.

 

[Sean Olive]

I can't imagine how one could possibly end up with a genuinely "level matched" comparison of headphones with different FR profiles - which is almost all of them as far as I can see.
My scepticism is very high here.

It's never perfect, but you can get close if the spectrum of the music is broadband and your loudness matching is based on such a signal. If the programs vary significantly in bandwidth you can do loudness matching separately for each program.

I actually think loudness matching is more important when the differences are very small like amplifiers and other electronic devices, particularly ABX tests where the goal is to identify one from the other. Usually small loudness differences are more noticeable than anything else and that is what people are using to identify.

But when there are significant differences between products in in terms of low frequency extension, resonances, colorations, spatial, distortion those things stand out from loudness differences, and trained listeners can identify the differences between loudness and coloration. The main purpose is to remove loudness differences as much as you can so that it doesn't with the reporting of other perceptual attributes.

 

[amirm]

I think this is something those advocating approximated EQ filters based on measurements of a single unit at a single placement are missing. Both their method and that of people like Oratory who average multiple units and placements result in 'smoothed' EQ curves, the difference being the smoothing in the latter is based on averaging actual measured unit and placement variation, whereas the former relies on effectively guesswork for unit variation, and a combination of eyeballing and (fallible) subjective audial judgement of placement variation with all the nuisance variable (e.g. cognitive bias) pitfalls that entails, diminishing the accuracy, reliability and generalisability of EQ profiles created in this way.

To the extent you have told us Oratory creates his EQ using his ears, then cognitive bias and other nuisance variables like him being in headphone business (and liking what they design) will loom large with him. So if you want controlled results, you don't have it in him.

As for "smoothing" the results, here are his AKG K371 filters:

Are you kidding me? Q of 0.71? Not 0.7, but 0.71??? You really think he could reliably tell the difference in a controlled test between Q of 0.71 on that high shelf and 0.7? Research shows our sensitivity there to be a few dBs, not one hundredth of a dB.

All of his filters have too much precision. Look at all the ones with gain of -1.9. Or 1.8. These are all like someone sticking a wet thumb in the air and saying the temperature is 33.1 degree!

These look mechanically created to me. I can't imagine him sitting there and futzing with 0.1 Q variations and arriving at anything valid. The amount of time it would take to develop such would be huge anyway.

Look at my EQ for example of perceptually created filters using measurements as guide:

See how there are no fractions on the Q factors? And how few filters there are?

I can guarantee you that my 3 and 4 dB corrections above are audible to all sighted or blind. The 1.5 dB one is tougher and you could leave it out if you like. There is no way you can say that about his filters as I explained above.

Are we to believe that with all their research, Harman created the K371 to need 9 filters to match their preference target as Oratory created? Doesn't pass the smell test, does it?

BTW, there are also side-effects from these filters in that not all implementations do what you see in the pretty graphs. Filters can have overshoot before and after that is not visible in the UI. This is another danger in auto-generating filters.

Bottom line is what I said: he is over filtering because he is chasing a mechanical target with measurements that are too variable to justify such. Averaging a few headphones doesn't help in that regard because each measurement is variable in itself. What helps is understanding the nature of variability and making proper judgement in developing filters as I try to do (using psychoacoustics and strict listening protocols).

 

[ninetylol]

Spoiler: post

To the extent you have told us Oratory creates his EQ using his ears, then cognitive bias and other nuisance variables like him being in headphone business (and liking what they design) will loom large with him. So if you want controlled results, you don't have it in him.

As for "smoothing" the results, here are his AKG K371 filters:

View attachment 109247

Are you kidding me? Q of 0.71? Not 0.7, but 0.71??? You really think he could reliably tell the difference in a controlled test between Q of 0.71 on that high shelf and 0.7? Research shows our sensitivity there to be a few dBs, not one hundredth of a dB.

All of his filters have too much precision. Look at all the ones with gain of -1.9. Or 1.8. These are all like someone sticking a wet thumb in the air and saying the temperature is 33.1 degree!

These look mechanically created to me. I can't imagine him sitting there and futzing with 0.1 Q variations and arriving at anything valid. The amount of time it would take to develop such would be huge anyway.

Look at my EQ for example of perceptually created filters using measurements as guide:

See how there are no fractions on the Q factors? And how few filters there are?

I can guarantee you that my 3 and 4 dB corrections above are audible to all sighted or blind. The 1.5 dB one is tougher and you could leave it out if you like. There is no way you can say that about his filters as I explained above.

Are we to believe that with all their research, Harman created the K371 to need 9 filters to match their preference target as Oratory created? Doesn't pass the smell test, does it?

BTW, there are also side-effects from these filters in that not all implementations do what you see in the pretty graphs. Filters can have overshoot before and after that is not visible in the UI. This is another danger in auto-generating filters.

Bottom line is what I said: he is over filtering because he is chasing a mechanical target with measurements that are too variable to justify such. Averaging a few headphones doesn't help in that regard because each measurement is variable in itself. What helps is understanding the nature of variability and making proper judgement in developing filters as I try to do (using psychoacoustics and strict listening protocols).

I never doubted his EQ settings are derived mathematicaly from the difference of the preference curve minus measurements.

Isnt this aproach the more objective based method instead of doing it by subjective ear?

 

[amirm]

I think it should also be considered that a user doesn't necessarily have to use all of Oratory's filters. He usually includes a broad low and high shelf filter to bring the overall tonality to the Harman curve (even stating that the low shelf can be adjusted to taste, which is consistent with Harman's bass preference findings). His other filters are then usually negative-gain peak filters (to avoid inadvertently adding to an unexpected peak, which would be more objectionable then an unintended dip). I think the best approach with his EQ profiles is to apply the shelf filters, then a couple of the main (lowest-Q, highest gain) peak filters, listen, then apply the rest in turn, one after the other, and if at any point the headphones sound worse to the listener (blind listening could even be done), just remove the last applied filter and leave it there. It doesn't have to be an all or nothing affair, and we shouldn't let the 'perfect' (exactly matching the Harman target) be the enemy of the good - bringing the frequency response on average closer to the target and so improving the headphone's sound for the majority of users, units and head placements, via averaging.

That doesn't work because his filters are fully interacting with each other. Look at the one I post again:

Look at the ones at 2700, 3400, 3800. They are all interfering with each other as they have significant gains. The moment you leave one behind, you are cooking up your own recipe devoid of whatever logic he used to develop his.

That is the drawback of creating such a complex filter recipe. You can't mess with it easily.

Even in my filters there are dependencies at times. I often put in a negative Q filter to limit the bandwidth of the previous filter. As such, the two filters come as a pair and are not meant to be used independently.

He could develop filters using a priority scheme and then you could do as you say. But he has not. You may get lucky that a subset works but that is not because you are correctly using his work.

If you want fewer filters, best use a recipe that starts that way.

 

[sweetchaos]

here are his AKG K371 filters:

Here's both Amir's and Oratory's PEQ filters.

AKG K371 - Amir:

Preamp: -4.5 dB
Filter 1: ON PK Fc 2200 Hz Gain 1.5 dB Q 1.0
Filter 2: ON PK Fc 6800 Hz Gain -3.0 dB Q 4.0
Filter 3: ON PK Fc 3915 Hz Gain 4.0 dB Q 4.0

AKG K371 - Oratory:

Preamp: -2.8 dB
Filter 1: ON PK Fc 170 Hz Gain -1.8 dB Q 1.2
Filter 2: ON PK Fc 1080 Hz Gain -1.0 dB Q 2.0
Filter 3: ON PK Fc 2700 Hz Gain -2.7 dB Q 2.3
Filter 4: ON PK Fc 3440 Hz Gain -2.5 dB Q 4.5
Filter 5: ON PK Fc 3800 Hz Gain 4.5 dB Q 1.0
Filter 6: ON PK Fc 5630 Hz Gain -3.8 dB Q 4.0
Filter 7: ON PK Fc 7300 Hz Gain -1.9 dB Q 7.0
Filter 8: ON PK Fc 8050 Hz Gain -1.9 dB Q 6.0
Filter 9: ON HS Fc 10000 Hz Gain 2.5 dB Q 0.71

Here they are both compared:

Enjoy!

 

[amirm]

I never doubted his EQ settings are derived mathematicaly from the difference of the preference curve - measurements.

Isnt exactly that the more objective based method instead of doing it by subjective ear?

I watched a youtube video of him where he said he doesn't use any tools to generate them and that he is doing it by listening. As I noted though, I think this is wrong.

As to objectively doing it, it is the problem I explained. We can get large variations in measurements even with one headphone. I can get bass response that varies by 5 or even 10 dB. Frequency of peaks and dips can change.

Furthermore, the target itself is rough. No room measurement looks like the target we use:

Above is two cups of the same headphone. Which is the "truth" to use for mechanical filter generation?

Machine generation would also be devoid of psychoacoustics. Would you want to correct for the wiggles around 5 to 10 kHz? Where would you draw the line and why?

By experience I have learned to not pull down the extra energy around 200 Hz. A mechanically generated one would spit out a filter for this.

Now, can a person build a smart tool that generates good filters as I do in my head based on measurements? Yes. But that is not what is in front of us. And I still would want to verify the results.