A Survey of the Utility of Headphone Measurements

Some very interesting words about headphone measurements will be appearing soon. But before we start on that, I'd like to survey InnerFidelity readers about their experience with headphone measurements.

The first poll is to get a feel for the utility of headphone measurements for enthusiasts.

The remaining polls intend to get a feel for how easy or hard it is to interpret measurements, and what you're looking for in measurements of the bass, mid-range, and treble areas of the measurements.

Polls are quite interesting to me, but they're also relatively limited. I'd love to hear specifics about your experiences relative to each poll in the comments. I'm particularly interested in hearing whether or not the Harman target response curve matches your particular listening preferences.

Should be an interesting exercise in light of upcoming posts.

Looking forward to your feedback!

Click here to go to the first poll.

COMMENTS
ADU's picture

I'm assuming these polls are all about the _raw_ frequency response, rather than compensated. Hopefully most here know the difference. :)

The Harman curve is close imo. But no cigar, because it doesn't do a good enough job of modeling the HATS resonances at approximately 2.7-3.5 kHz, 9-10 kHz, and 15-16 kHz in the treble.

Some of the transitions in the bass and mid-range could be a little smoother as well imo.

Tyll Hertsens's picture
Yes, raw plots.
Mad_Economist's picture

In that case, I'm afraid I've contributed to your margin of error a bit. I missed the tip-off of the raw plots in the post announcing this poll, and answered as though the questions addressed the compensated plots. Whoops!

zobel's picture

1) sub bass = 10Hz - 25Hz
2) low bass = 25Hz - 50Hz
3) bass = 10Hz - 200Hz
4) upper bass = 100Hz - 200Hz
5) low mids = 200Hz - 400Hz
6) midrange = 200Hz - 1.6kHz
7) upper mids = 800Hz - 1.6kHz
8) low treble = 1.6kHz- 3kHz
9) treble = 1.6kHz- 20kHz
10) high treble = 10kHz - 20kHz

Tyll Hertsens's picture
I'd put the 1.6kHz at 1.2kHz...but that's splitting hairs.
LytleSound's picture

I think that you should consider dropping the 1.6 kHz to 1.25 kHz. Also why do you have overlapping ranges of bass and upper bass and low mids? Otherwise you have adjacent bands for each label. except that treble isn't 1.2 kHz to 20 kHz, but it is 3 kHz to 10 kHz. Remember that there are NO fundamentals in music above 4 kHz. The harmonics above that frequency are seriously reduced. The exception to the rule is from certain drum kits where there is significant energy above 8 kHz, which would fall into high treble.

zobel's picture

You might glance again at those ranges.

I have Bass (2) @ 10- 200Hz
....Mids (6) @ 200Hz - 1.6kHz ......and treble (9) @ 1.6 - 20kHz.

Nothing overlapping there. Just researching the general consensus from lots of sources, the treble starts about there. Tyll and I are so close here that we would picking nits on the start of the treble range. I think it is a good idea for Tyll to publish this info under RESOURCES above, so that everyone can be on the same page when we talk frequency ranges.

ivanhagberg's picture

For quite some time now statements like "In spite of its measurements it sounds terrific ..." and "... looking at its measurements it is surprising how good it sounds!" and similar have been around and thus give at hand that there are plenty more involved than what the current curves can provide to us to describe how headphones REALLY sound like.

Obviously the measured frequency response is somehow accurately representative, but really, how much is it responsible for to the final sound experience. There seems to be a LOT of extra factors involved in what characterize how headphones REALLY sound like. If one compares different singers singing the same tune, they can keep the frequency quite identically, but how different do they not sound? So many other acoustic artifacts are coloring their voices and how they soound to our ears. I feel that the Harman curve is only there for getting a balanced sound volume over the spectrum, nothing more. But resolution, sound stage, instrument placements, handling standing waves, damping materials, the chamber shape etc, seem to have greater bearing on the sound.

To me measurements might be able to show if the headphone has a particularly pronounced character, a serious flaw or being broke or not, but other wise ..., no. You must LISTEN yourself!

The best reviewer to me is the one that can portray and convey his listening experience in an understandable way and primarily and consistently draws conclusions from his listening sessions and his experience with OTHERvheadphones and less from his curves. Of course he must also use very good source material, otherwise defects in the recordings can end up as comparison material and erroneously give an upper hand to a poor headphone that is not able to reproduce that error but hides it and the quality headphone gets a poor review since it sounds "bad" at that point.

ADU's picture

Thanks, Ivan.

Long time listener's picture

If the distortion levels are too high, I won't like them. If the bass falls off below 80 or 100 Hz, I won't like them. If the levels are too high at 3-5 kHz, I won't like them. A response curve in which bass, midrange, and treble blend smoothly into each other will sound better than otherwise (AKG 518LE still sounds pretty good to my ears, and fits this description). So measurements will tell me if a headphone is junk or not.

I find that IEMs MUST have a much higher low- and mid-bass response to sound SUBJECTIVELY as bassy as an on-ear or over-the-ear headphone with a more flat response. So I don't think the Harman curve applies well to IEMs.

Tyll Hertsens's picture
Sean Olive of Harman once told me that they intend to work on determining an IEM curve by working through the subjective listening panels like he did with over-ears. I suspect he is aware that there indeed might be a different curve, and is open-minded enough to test them as a separate case.
24bitbob's picture

I find iem's too variable depending on the fit you manage to achieve. I find that both for custom and universal. When they're good they can be scary good, but the difference of half a mosquitos ball hair can knock them out of sorts. Maybe it's my ears?

Great article for over the ear headphones though, and I'm interested to see mention of EQ when it comes.

gibtg's picture

I think the debate for the usefulness of measurements is a waste of time. Without measurements there are only opinions, which are worthless in terms of the scientific method...

And if we don't want any science involved there's no point in analyzing the headphones to any of the depth that is usually done here.

ultrabike's picture

Measurements are not the scientific method.

I do however believe that measurements are of tremendous importance in audio performance evaluation. Specially considering that many audio releases and accessible song re-masters these days are utter and shameful pieces of shit, and mask the performance of good and horrid audio reproduction equipment. Which in turn makes audio evaluation by ear difficult unless you have trained ears and decent source material with which to evaluate.

Furthermore, I believe that a well established, credible, and highly accurate audio evaluation approach or approaches, would force companies in producing decent audio reproduction gear... and better music. At least in an ideal world.

Beagle's picture

Science does not always sound good or bad to the ear. If science was/is the determining factor, one would hardly ever enjoy a live musical event in a concert venue.

ultrabike's picture

IMO live musical events in a concert venue, unless we are talking about classical music or orchestras in a well thought of theater/opera location, tend to sound like shit.

Last time I went to a fair in Costa Mesa, CA with my wife to listen to some pop band we had a great time, ate some awesome junk food, and overall enjoyed the atmosphere. But sound quality was horrid.

elfary's picture

I am as well of the opinion that in ear monitors need a bit more bass energy than that of harman curve.

Lots of times i have found that some in ears whose measurements look way too hot in the bass region actually feel almost right (Westone W30 comes to mind) when heard.

Also it's important to point out than fr requirements change depending on sound pressure levels. And on in ear monitors depending on background noise as well.

Hence i think more than just one idea curve there should be two at least (headphone, iems) even twice of both depending on listening levels.

Stefraki's picture

I find them useful, and can use them to get a good idea of whether I will like a set of headphones or not in most instances.

This only comes about from experience of listening to a lot of headphones and comparing their charts against my subjective likes and dislikes.

I can know, for instance, that I am very sensitive to distortion and won't like a set of headphones that doesn't have reasonably low distortion. I also know that there is much more tolerance for distortion below 100hz.

I can recognise a frequency response that is not going to be to my tastes—such as an excessively 'V-shaped' presentation or excessive lower mids, and have come to recognise weird quirks I actually enjoy—such as a little extra energy around the 1-2khz area.

But I only get that knowledge through subjective experience and applying it to the objective data. Without enough reference points I don't think anyone could tell very much from them.

ADU's picture

...And I agree with much of the above.

jander5s's picture

I noticed you focused on frequency response. I do find the time domain plots also quite useful.

ultrabike's picture

Measurements can tell us quite a bit about how a headphone might sound. But how much depends on:

1) Type of headphone.
2) How well we understand the measurement(s).
3) How well we can correlate particular measurements with actual performance.
4) How well we understand what we want to get from a headphone.

In particular, I like flat frequency response (when measured independent of ear shape and on full size cans), and minimal distortion across the audio frequency range.

I also believe that IEMs will have more difficulty in having a universal reference response, i.e. person ear characteristics independence.

But I do think that full size headphones can be somewhat person ear characteristics independent and therefore more universal in performance.

zobel's picture

I wonder what you would consider a flat frequency response curve (or graph of SPL/frequency) would look like when measuring cans on the dummy head? Obviously a flat line would sound awful. Which of Tyll's raw frequency response curves are closest to flat?

I agree that IEMs are harder to pin down an average for, mainly due to the larger variations in ear canal shape and volume, over the variations found averaging the much bigger volume encountered by the drivers in over the ear cans.

I think waterfall plots are very helpful. The cumulative spectral decay plots show resonances superbly, pinpointing their frequency and duration in an easy to read format. I understand that this plot can be derived from the impulse (step) response.

The THD+NOISE/Frequency graph is marginally helpful, don't you think? Since we can't sort out what order of harmonic distortion is there, and how much resonances are showing up in that chart, not to mention the trucks going by at Tyll's place adding to the measuremnt, its tough to get reliability in correlating sound to that set of lines, don't you think?

I,too find IEMs to lack bass for the most part, but mainly find them very uncomfortable, and they seem to make ear wax production go into hyperdrive. I realize everyone's results vary with those.

ultrabike's picture

A raw flat frequency response should sound pretty decent, if the measurement procedure does not include dummy heads, keeps seal, and minimizes internal modes/reflections due to the coupler.

A dummy head will not yeld flat measurements, and different dummy heads will yield different responses. Measurements with dummy heads should be compensated, and each different dummy head may require it's own individual compensation.

For waterfall plots, they are meaningless if using a dummy head, unless the impulse response measurement is compensated in the time domain to account for dummy head. If using a coupler as describe above, no compensation may be necessary.

THD and noise IMO is not irrelevant. In a way, it shows that the headphone produces sound on it's own accord that is not faithful to the source. It also pin points to some extent which frequency ranges deteriorate the most due to the physical limitations of the transducer.

In general, you do need to perform your measurements in a quiet environment, since you are measuring sound pressure levels. AFAIK Tyll measures inside an an-echoic chamber.

IEMs IMO are great when you want sound isolation. But they can be uncomfortable, and I cannot run with them due to occlusion effects.

zobel's picture

Have you been measuring cans as you described? If a cumulative spectral decay plot isn't possible through a dummy head, then how is the impulse response possible there? Wouldn't the waterfall start at the raw curve? There really isn't any time domain distortion, just resonances and comb filtering, right? I do agree that it would be much easier to see if the raw data produced a flatter line.

I think I have to agree with you about the problem of measuring SPLs at eardrums. I've thought that since there are so many resonances produced that the brain filters out, we don't get a real representation of the actual perceived sound when measuring there, not to mention as you pointed out, the differences between heads that would need compensation for. What do you use? I don't measure cans, just speakers.

Do you think there is any chance that all this SPL/frequency data Tyll has acquired could be correlated to a model that represents flat using his raw data? The Canadian curve is just that, a rough, smoothed out general pattern of what raw data at the eardrum looks like when a room compensated flat response is perceived. I say room compensated because, as you know, an 'average good listening room' is the model used by playing flat (anechoic) loudspeakers in it. Nobody wants to hear the anechoic response of speakers. Flat speakers there sound lifeless, and lacking in bass, so they model a good listening room.

Therefore, doesn't the target response of the headphones need to be the averaged in room response of those flat speakers? If we could find a way to measure the headphones in an anechoic chamber, wouldn't that chamber have to offer the same acoustic compliance as the driver would encounter on someones head? That means it would have to be a very small acoustic chamber, and by definition not anechoic. That is the catch. Whatever the cans are connected to will influence hugely their response. Due to their dependence on being on (or in) an ear, how do you model a more representative chamber, with a proper averaged representational loading, but diminished comb filtering and resonance? What has worked so far, if anything?

I didn't say I thought THD & NOISE / frequency was irrelevant. I think it is useful, but not anywhere as accurate an indication of perceived audio quality as knowing the nature of the distortion spectrum would indicate. IM distortion would be another good test, don't you think? Tyll was mentioning the problems he is having with street noise contamination. The relatively tiny proportion of noise being measured is easily contaminated by the slightest outside sound, and he has been fighting that with his situation there.

IEMs.. gotta pass on them, though some folks are happy with them. I can't run with them either, not because of excessive isolation, but because of two hip replacements, a spinal laminectomy, and too many miles on the associated parts (insert appropriate emoticon).

ultrabike's picture

This is long. Sorry about that.

"Have you been measuring cans as you described?"

Yup.

"If a cumulative spectral decay plot isn't possible through a dummy head, then how is the impulse response possible there? Wouldn't the waterfall start at the raw curve?"

A CSD through a dummy head is possible, but the impulse response obtained through it has to be compensated in the time domain before generating the CSD. The waterfall should be obtained from the compensated impulse response. One could generate the equivalent impulse response of the compensation assuming the compensation is minimum phase and convolve the two impulse responses. Then obtain the CSD from the result using a high resolution window.

"There really isn't any time domain distortion, just resonances and comb filtering, right?"

Nope. There is time domain distortion corresponding to the frequency domain distortion. It may be possible to visualize this by thinking of non-linear distortion as a separate input to an ideal transducer.

"I've thought that since there are so many resonances produced that the brain filters out, we don't get a real representation of the actual perceived sound when measuring there, not to mention as you pointed out, the differences between heads that would need compensation for. What do you use? I don't measure cans, just speakers."

I try to use a similar approach to what is done with speakers. I believe the problem is that speakers drive a larger volume than headphones. For full size headphones, the volume in the ear drum may be negligible compared to the volume between the headphone and the ears. But it may be hard to measure a headphone driving the required volume of air and avoid reflections as when using an anechoic chamber for speakers.

"Do you think there is any chance that all this SPL/frequency data Tyll has acquired could be correlated to a model that represents flat using his raw data?"

Yes, using the appropriate compensation curve corresponding to the head-torso model that Tyll uses.

"Flat speakers there sound lifeless, and lacking in bass, so they model a good listening room."

An anechoic chamber will likely have negligible reverberation. IMO it should be able to reproduce the recorded reverberation well, if such exists in the recording.

If you are using room treatment, it seems to me you are trying to approach an anechoic chamber. It may not be what you want to do if you like your room effects though.

"Therefore, doesn't the target response of the headphones need to be the averaged in room response of those flat speakers?"

Not if you want a high fidelity reproduction, IMO.

"If we could find a way to measure the headphones in an anechoic chamber, wouldn't that chamber have to offer the same acoustic compliance as the driver would encounter on someones head?"

Not IMO.

"That means it would have to be a very small acoustic chamber, and by definition not anechoic."

Nope. A small acoustic chamber may be anechoic if it does not produce echo. That maybe a bit difficult to do relative to what can be done for speakers.

"What has worked so far, if anything?"

Well, bass response may not be affected that much. You can try different material for the mids and treble and compare results between open air measurements and measurements with the coupler. This was sort of done to some extent by JA in Stereophile with some Grado cans.

"I didn't say I thought THD & NOISE / frequency was irrelevant. I think it is useful, but not anywhere as accurate an indication of perceived audio quality as knowing the nature of the distortion spectrum would indicate. IM distortion would be another good test, don't you think?"

If what you are looking for is a pin point accurate idea of what distortion is going to do, that would be kind of hard. Because that would probably largely depend on the music you are listening to. But IMO, you can get a rough idea. IMD measurements IMO wouldn't help much in pin pointing distortion effects, but it can give you an idea of the limitations of a driver.

"The relatively tiny proportion of noise being measured is easily contaminated by the slightest outside sound, and he has been fighting that with his situation there."

If you average multiple measurements, random noise should be minimized. Trucks and more correlated noise might be a problem. So you do your measurements when things are quiet. If you have an anechoic chamber on top of it then you proly will not have that many issues, but I could be wrong.

zobel's picture

Just picking you mind, thanks for your thoughts.
I don't want to look for differences of opinion, here, since that is just that, opinion, but just this, the free air response of cans must lack bass, as you know, compared to sealing next to ear. A small chamber will always increase bass coupling with the driver. Small chambers can't be anechoic, since standing waves occur there, and the entire chamber is pressurized by the driver, and boost bass IMO.
Its interesting that anyone would want their listening space to be anything like anechoic. You can approach this by putting your speakers outside on tall stands away from any reflective surfaces. No bass. Dead sounding. That is why the target response is for good speakers in a good room.

You are right, even though the drivers in cans are minimum phase, resonances in the cans might add to the SPL. Assuming, though, that a very low measured THD+Noise would indicate that this type of noise is not an issue, and the response is basically minimum phase for all practical purposes, correct?

The most recent post above here is interesting isn't it? How do your measurements compare to others you see? What do you use for your measuring set-up?

There will never be an accurate one-size-fits-all measurement system for headphone frequency response IMO. Considering all the known and unknown variables involved on all those individual ears, heads and brains, the objective opinions of many, and maybe one of a trusted similar set of ears on an honest experienced reviewer are what gives us the most information now. And, no dummy head will tell you about the comfort and fit.

By the way, I have enjoyed the objective reviews you have done, and find I mostly agree with you. Have you heard the new Ultrabuck Audeze LCD-4? I haven't, nor will I bother with them. How about the ...oh never mind :)

ultrabike's picture

If a small chamber could ideally absorb standing waves and reflections perfectly while maintaining proper air pressure, then you would get anechoic behavior. I'm not certain this is an impossibility.

As far as speakers, many full size speakers are measured in anechoic chambers, and are able to reproduce sub-bass frequencies. It makes sense to me however that ability to reproduce low frequencies is linked to the volume of the driven media.

Furthermore, drivers are not necessarily minimum phase. Nor is THD or noise minimum phase. THD and noise are a bit of a different animal.

As far a my measurement setup, I use a flat sealing surface with a bit of foam inside to absorb as much of the reflections as possible. It is not perfect IMO, but I'm happy with the results. And it's cheap and easy to make.

IMO, there should be (eventually) an accurate one-size-fits-all measurement system for certain kind of cans. It's just a matter of time and willingness.

I have heard about the Audeze LCD-4. Most Audeze cans have not impressed me that much. I'm pretty happy with my HD600s.

zobel's picture

I meant to say the 'subjective' reviews you do. I still dig my HD600 too....I just need a
better low bass, a tad more air in the treble, and a bit more of a dynamic overall presentation...... not too picky huh? ;)

Good to visit here!

Aufdemaury deus ex machina's picture

The Akg k872 (closed back version of k812) came out, looks like some driver changes also occurred,
probably won't fix much of it's initial open back predecessors problems, but it would be intriguing if you investigated what it's all about. I personally noticed measurements of Tesla magnet drivers in closed back measure quite a bit better than open backs teas magnet headphones, so it might make a difference, though thats just my speculation and observations.

Thanks Tyll

Aufdemaury deus ex machina's picture

They came out with a new line of some sub 300$ close back headphones, looks a bit curios as well

Cats_Paw's picture

I think measurements are required because a certain part of the frequency response might be off by a decent amount and we might not notice it long after purchase (because a particular set of songs never hit in that area so we never noticed or what is even worse, it sounded great in some songs because it counteracted a problem in the song itself, but sounded terrible in other material).

Personally, I like slightly recessed treble, and I know I like it because I was able to look at measurements of HPs I liked.

LytleSound's picture

Sean Olive is a blessing for this field. Until now we have been using the diffuse field response corrected for the insert. If Sean gives us a curve at the TM that we can use with couplers as he did for circumaurals those who use IEMs will be better for it.

Bob Katz's picture

...between your top choice and the second. My answer would be somewhere between your top choice and the second one. I also posted some suggestions for future collaboration between measurement scientists and the practical results in reply to Audeze's letter that you published. We are on the crossroads of greatness!

X