The Science of Headphone Sound: Rant, Ideas, & Discussion

Apologies if there is already a discussion specifically on this subject, but I have a few things I would like to hear thoughts on and to get off my chest.

I think like many of you, I like science and technology. Like many of you as well, I did not end up as a researcher at a major university; however, my thirst for knowledge, technology and basic comprehension is there. I feel like the vast majority of audiophiles that end of up here or looking at measurements in general rather than talking about the auditory merits of 8 vs 16 core headphone cables fit in a similar category. I also want to add that I have never measured a headphone, so wanted to make sure you all knew to take my opinion and ideas with a grain of salt.
That being said and out of the way, I have some rants that I am going to organize like the scientist I should’ve been:

  1. Errors related to accuracy are typically systematic and uncertainties around precision are usually random. Repeating an experiment improves precision, but usually doesn’t improve accuracy… however, if you change the experiment or increase the variables being measured, accuracy will increase. I say this to relate to taking measurements of headphones and specifically to that of the method used by Resolve, Crin and others VS that of ASR. I know @Resolve has said and I hope he chimes in that he takes multiple measurements with the headphones in at different placements of seal etc… ASR is the complete opposite as Amir says in bold on every one of his reviews: “Headphone measurements by definition are approximate and variable so don’t be surprised if other measurements even if performed with the same fixtures as mine, differ in end results. Protocols vary such as headband pressure and averaging (which I don’t do).” Which is his way of saying, I take one and go… My rant is just that I appreciate Resolves take here. I think it will more “scientifically” prove the accurate and most likely representation of a Headphone’s FR across the board. Ideally, we would be able to do that to multiple samples of the same headphone because it seems many times FR varies depending on unit.
  2. Shilling for respect, power, or just reputation in general is just as bad as shilling for money. Not naming names, but in a world where snake oil flows morse so than many others, I value those who have the integrity to say they may be wrong or it’s just their opinion more so than those that deal with absolutes or saying they are right and no one else’s opinion matters.
  3. I think Resolve is on to something with trying to figure out what detail, speed, separation, and slam really are and how to measure them. Again, to slam, get it, on ASR, the idea that those things must be represented in a FR is such a close-minded approach. To call yourself a scientist and say such an absolute is nearly blasphemy. Obviously, headphone research, even at Dr. Olive’s level, isn’t world changing, but the history of science and discovery at ALL levels is full of tales of thinking outside of the confines of what you know. I think that if ways to observe and record data that proves varying levels of detail, slam and speed are discovered, it will be much easier to review and relate the difference between “cheap” and “expensive” headphones. It will also then be easier to differentiate those things within both categories.
  4. I do not have the capacity or instruments needed as mentioned above, but I would like to hear people’s thoughts and possibly see a few methods investigated if they haven’t already been…
  • a. For Detail and separation… I wonder if there would be a way to take an audio spectrum analysis while a headphone is on the Gras or similar device of a particular sample of music that is considered to have high levels of recorded detail and layers of sound (very busy so to say) and compare across multiple headphones. Things we may see would be the presence or lack thereof those details. Would be easy to do a “control” of the sample by just looking at the audio file itself digitally to see where we would expect to see those details.
  • b. For speed… I think there may also be a way to view the spectral analysis for this while the headphone is on the Gras when comparing to that of a similar control as above. Or possibly using the waterfall graphs when compared to particular passages.
  • c. For Slam… I really believe it has to do with Air Pressure/volume/force so coming up with a way to measure the changes in air pressure could be interesting. I know that Gamers Nexus just got a pretty crazy PC fan wind tunnel that could possible measure such small variances. Maybe it could be done much easier than that. I don’t think it’s too hard to deduce while one may perceive the he6 to slam harder than the h400se… the diaphragm and magnets are way bigger thus most likely being able to move more air.

Many times though the experiences we have that we call detail, slam, separation, and speed are usually comparatively to other headphones, so making sure to do just that would be imperative.

  • d. I would love to see FR graphs of the same headphone (the control) being powered by various amps… I think this could be a very simple way to show the perceived differences between gear. Perhaps just something to start as simple as HD6xx measured on a clean solid state versus a tube amp… I am surprised I have never seen this experiment, but I know I have seen it on Speakers/cables as GR research has done similar videos comparing FR of speakers

I will get off my soap box and go back to enjoying my music, but would love to hear everyone’s thoughts!!!

5 Likes

My 1/50th of a dollar here. I think the Objective, Subjective, Dejected thread covers some of this. Importantly, it’s one thing to measure and a different thing to hear due to physiology, anatomy, and psycho-acoustic considerations. This is not to say that there might not be some reason to have more, better, possibly more objective measurements,

I’ve thought about separation and how to measure that. Unfortunately most sonic devices that are good at measuring distance seem to be ultrasonic - think of the backup sensors on a car bumper.

I would think that you’d need to design a stage with specific micrphone placement and know exactly what distances exist between identifiable sources. This would not necessarily be music. Then you’d have to do some sort of real-time analysis of sound reproduced by headphones. Then I don’t know what. As a scientist, I might be a member of the tubular earth society.

4 Likes

Yeah there is truth to this, but I think the bigger obstacle is that getting all the potential variables in behavior - even just with FR - is extremely time consuming. One of the criticisms that a number of us have had with his methods is that he doesn’t approach the measurements with sufficient rigor to get it right consistently, but there’s also the reality that doing so would take quite a bit more time and require additional methods.

It can sometimes take me the better part of a day to get measured results out because of the need to check that seating/positioning matches perfectly with how I hear it, or I need to run the in-ear mic test to check bass coupling and so on. I think his response would be that with a 45CA he shouldn’t need to do that, but the reality is that even a 45CA isn’t an actual human head, it’s just two ear and cheek sims on either side of a stand, and it also lacks the side of the head contours that would guarantee representative coupling. So you might as well test with an in-ear mic as well to get at least one real human datapoint for bass. And I’d argue that with some headphones, it’s actually more difficult to get a representative seal on a 45CA than on the 43AG we use.

Another element to this is that he seems to weight imperceptible qualities (like THD) more highly than the obviously perceptible ones like FR. I think this is anchored to a headphone’s ability to EQ, since EQ is basically required on most headphones (certainly all open-backs) to achieve the desired target. And even if the THD is below the audible threshold, that seems to count for more than the default FR’s adherence to the target. In some respects, I can understand the merits of this approach, if you think target adherence is the most important thing, and you know you’re going to EQ. Not saying I agree with it, but I think that might be a way of making sense of it.

We actually do this all the time, as in measuring from two different devices for the same headphone, it’s just not commonly published because there’s no difference. Unless there’s some unique high output impedance situation, with modern solid state devices the norm would be that you don’t see any difference (for solid state sources). I think that’s also aided by the trend in SS devices to all aim for similar measured results, for better or worse.

There’s a common notion that the source should add literally nothing to the sound, and that seems to be a target for many manufacturers. What’s cool to see is that some are trying to think a bit more outside the box on that stuff as well, like some of my conversations with certain engineers suggests they’re deliberately trying to conserve qualities that may not be as desirable in measurements but have an interesting effect. In any case, I’m often reminded that this hobby is a blend of art and science, and navigating that is never going to be straightforward.

8 Likes

There are merits to both approaches. An averaging approach eliminates the scenario where an outlier position that’s measured doesn’t reflect “normal” on-head behavior; a single-position measurement avoids the “fuzzing” of positional details together. IMO it would be more ideal to have a plot of the average, variation at some given confidence for the number of placements used, and “worst-case” for each frequency. Keith Howard does something similar to that with the confidence limits on Headphonetestlab.co.uk


although personally I’d prefer to plot variation vs. frequency relative to the mean.

It’s hard to say that you’re wrong, publicly - particularly with how contentious this hobby is. That doesn’t excuse us from doing it, but with the amount of hostility that’s thrown around, particularly at controversial figures like Amir, I can see how it’s hard to let go of defensiveness and be honest about errors (and I say that as someone who’s had the frustrating experience of trying to convince Amir that he’s in error).

One thing to keep in mind with this concept is that, for almost anyone in this hobby, there are a least some descriptors that you think are absolute hooey - it might be “slam”, it might be “detail”, it might br “PRaT”, or the “airiness” of a cable, or how much “cleaner” a ziplocked baggie of pebbles makes things, but all of us are drawing a line somewhere, and that means we all have some bar that things have to clear. For Dr. Olive, that’s “have some evidence that it’s a quantifiable phenomenon that’s affecting listener preferences”, nearest I can tell - and with his work having shown that, at minimum, frequency response is the overwhelming majority of what’s impacting listener preference in headphones, you can see why he’d be skeptical of any supposedly separate factors.

You can, of course, record the waveform of a song or section of it playing through a headphone, using an ear simulator. From there, you can analyze it via a few methods - several of them are built into Paul Kane’s lovely DeltaWave application. You will see differences between the original and the recorded reproduction, of course - some of them are due to environmental noise that the microphone will capture (which cannot be kept constant), some are due to inevitable issues with aligning the waveforms, and some, of course, are due to the linear and nonlinear distortions of the headphones themselves.

I will note that generally music produces less significant nonlinear distortion than sine waves for a given peak level, due to its higher crest factor (music is, almost always, quieter than sine waves).

For the linear distortions…well, that’s frequency response - and it behaves the same regardless of the input.

Speed and bandwidth are inherently connected - you can see the “speed” of a headphone by looking at its high-frequency extension, and with 20khz bandlimited content, there is a maximum “speed” to any signal in your recording.

CSD waterfall graphs have the double disadvantage of being very sensitive to parameterization and not giving additional information about minimum phase systems vs. magnitude frequency response. Some dialogue and plots here.

Well, pressure is definitely pertinent to what we hear - it’s the P in SPL! A headphone which is incapable of generating sufficient air displacement is going to have a proportionately reduced low frequency response, although when it is tightly coupled to the head, the displacement required for a given eardrum pressure becomes constant below the point where waves are very long relative to the volume’s dimensions.

Also, as an aside, the HE400SE, to my knowledge, has the same diaphragm size as the HE6 - the magnet arrays and trace patterns are different though.

As @Resolve said, this can be - and is - done, but there’s nothing interesting to see here outside of impedance interactions, and those tend to be a lot less interesting for headphones (which are fairly predictable loads dominated by their DC resistance) than for multi-way speakers (which have the added fun of crossovers, and much lower coil resistance).

4 Likes