MiniDSP H.E.A.R.S headphone measurement jig

I like to measure stuff. Does that place me on one side of the long-standing objectivist vs subjectivist debate in high fidelity?

Well, I’ll leave that to you to judge. But here’s what I do think: subjective impressions of audio gear are vitally important because they can identify aspects of performance that aren’t apparent in measurements. But pure subjectivity, untethered to any sense of objectivity (or, arguably, reason) can lead one to thinking that placing a piece of cardboard under one leg of your listening couch improves the sound.*

The main reason I measure stuff is that it helps me calibrate my own hearing. If I think something sounds warm, then I’d expect to see a modest boost in the mid-bass response. If there isn’t one, what is it that I’m hearing? Maybe I should come back the next day and listen again. Does something sound way too bright? This should be reflected in a measurable rising response in the treble.

If it turns out that my ears are working properly on these basic, measurable things, then perhaps I’m permitted to have more confidence in less measurable things, like the amount of detail I can perceived in a playback system.

That brings us to headphones, and how ridiculously difficult it is to interpret the measurements you may make of their sound.

Which is what I’m starting to do. (The headphone measurement jig is a relatively cheap device –professional models cost tens of thousands of dollars. Still, the MiniDSP H.E.A.R.S seems to provide fairly representative results and each one is provided with a calibration file to ensure microphone accuracy.)

What is a flat frequency response?

The most obvious marker of some item’s sound character is its frequency response. Particularly the tonal balance between the bass, midrange and treble. If I’m measuring most devices, what I want to see is a “flat” frequency response. By which I, and everyone else who does this stuff, mean that the line of the graph showing the output level at different frequencies is flat across its range. Here, for example, is the kind of frequency response I like to see from a headphone amplifier – in this case, the Topping A30Pro balanced headphone amplifier (see also our Topping A30 Pro review):

Topping A30Pro balanced headphone amplifier frequency response

As you can see, there is virtually zero variation across the whole audible bandwidth. If anyone tells you that this amp sounds “warm” (ie. it has a gentle bass boost) or has anything other than a totally neutral tonal balance, they’re either mishearing things, or have somehow coupled the device to totally inappropriate equipment. And I doubt the latter explanation.

Electronics are easy. Anything that interfaces between electronics and mechanics is much harder. So loudspeakers are hard to measure because what they do is in part what they do, and in part how they interact with your room. And that is due both to the characteristics of your room and their position within it.

Still, an ideal loudspeaker in an ideal room will produce a flat frequency response. There is no such thing, but we can judge to some extent deviations from ideal performance by the departures from the frequency response.

(And, yes, I know that various subjectively “ideal” non-flat, typically slightly diminishing, frequency response curves have been developed. But I’m sticking with flat.)

And all of that is a prelude to saying:

You do not want a flat frequency response in your headphones!

Let’s say that you entice a string quartet to come into your listening room and play you a rather nice Schubert piece. And you also have some fine engineers record the performance. You’d like the recording, when replayed, to sound as close to identical to the original performance as possible. One part of that would be a perfectly flat frequency response curve.

Now, let’s consider what happens when that string quartet is actually playing in your room. The sounds emerge from their instruments, bounce around your room from the various surfaces, and enter your ears. As it happens, you also are a surface. In fact, you are many rather complex surfaces. Your body modifies the sound, as does your head. And so do the various elements of your ears. Evolution didn’t stick those weird swirls in your ears for the LOLZ. They help adjust the sound so that you can do things like tell what direction it’s coming from (the better to keep from being eaten by that stalking beast). Your ears aren’t laboratory instruments. They are intensely practice devices developed over tens of millions of years to help keep you alive (at least long enough to reproduce).

So, what strikes your eardrums (and cilia and whatnot) when you’re listening to music is not the original sound from the quartet (nor from the speakers if you’re listening to the reproduced version), but the sound modified by all those interactions with your body and with your ears.

And studies have shown that all those modifications change the tonal balance of the sound enormously.

But what about when you’re listening with headphones? Many of those modifications are no longer happening. The sound isn’t bouncing from your torso. It isn’t finding its way around your head, reaching one ear slightly ahead of the other. Indeed, there’s effectively no bleed from one ear to the other.

Even the swirls in the outer part of the ear exert reduced influence on the sound, depending on the design of the headphones. If you’re using in-ear monitors, they are completely excluded from any influence on the sound.

Now, it could be that these influences on the frequency response could be minor, maybe a decibel or two. Unfortunately, that’s not the case. Extensive research suggests that the effects amount to a variation of around 20dB in total. And that’s enormous.

If a set of headphones had a completely flat frequency response, they would sound terrible. That research tends to show that a natural sound from headphones is achieved it the output is increased by in excess of ten decibels at around 4kHz, while pulled back by more than five decibels at around 11kHz to 13kHz. Headphones with a flat frequency response will simply sound wrong.

So quality headphone makers engineer their headphones to accommodate those adjustments.

It’s not as easy as that, though

All that’s fine as far as it goes, but the research is necessarily about how things work on average. Researchers use multiple listeners to assess the various adjustments to tonal balance and come up with an average figure. But you and I and everyone else hear things a little differently. Our bodies and heads and ears are different shapes.

So there’s tremendous individual variation in how we hear things from headphones. Which is why the headphones I love may not be the ones that you find satisfying. And which is why I always insist that you should audition quality headphones before buying.

Headphone making is as much an art as it is a science. Yes, science informs the whole enterprise, but the final tonal balance is a judgement exercised by the headphone maker. So, where do they put those peaks and troughs in the frequency balance? They’ll be roughly in the same place, but with variations, sometimes subtle, sometime obvious. And without resorting to digital signal processors, there are limits in the amount of tuning possible with headphones.

And even today further research is revealing additional dips and peaks in the frequency response that tend, on average, to make the sound seem more natural.

Let’s look at why I’m bringing all this up

Here is the “raw” frequency response measurement of a pair of Sennheiser HD 660S headphones that I measured.

Sennheiser HD 660S uncompensated frequency response graph

As you can see, they are beautifully balanced from around 80 hertz up to around 3kHz. But above that, they go through the roof, with a peak of about 20dB at a little above 4kHz. Above that they come down to a reasonable level and then wobble around quite wildly in the treble.

The Sennheiser HD 660S headphones are generally regarded as very good mid-priced headphones, right on the threshold of high-end audio. I certainly enjoy listening to music with them. But wouldn’t that enormous peak slightly above 4kHz make them an uncomfortable listen? No, because that compensates for the behaviour of our ears.

Obviously, such graphs are hard to interpret, what with those wobbles and whatnot. Fortunately our researchers have developed “compensation curves”. If these are applied to the measurements, the closer the “compensated” frequency response curve is to a flat line, the better. Here are the same headphones measured with the compensation applied:

Sennheiser HD 660S compensated frequency response graph

The treble is clearly much closer to an even balance with this compensation, although it still wobbles around alarmingly across a range of about a dozen decibels. But those wobbles are due to a combination in physical limitations in tuning, and artistic judgement by the headphone makers as to what they reckon sounds best.

And in the case of the Sennheiser HD 660S headphones, they’ve done, in my opinion, a pretty decent reckoning.

Now, from time to time I’ll be including headphone measurements in my reviews, and looked at naively, they’re going to look horrible. But I’ll be using the compensation curves, so they won’t look quite as horrible as the raw measurements. Still, pretty horrible. Just remember, at least with quality headphone brands, all that variable stuff in the treble reflects a judgement by the headphone maker on what things should sound like.

The bass end of the measurement, however, is far more representative of what bass you actually hear. That can be very useful information.


So what does all this mean? First, if you see a frequency response measurement of headphones, look to see how it was conducted. Is the measurement compensated for the generality of tonal effects revealed by research?

And then, I’d suggest, largely ignore the treble region. If you’re confident that appropriate compensation was applied, then you can get a sense of overall balance. However, I’d place a lot more weight on listening impressions from the reviewer.

The midrange and bass, though, are a different matter. The response graph can give you a very good sense of tonal balance there, and particularly reveal how balanced the deeper bass is.

If you see a large hump in the bass, that might make for a fun sound but not for a musically accurate one. And if you like super deep bass, look for that line to go as far to the left as possible without dropping down too far in level.

But, as always, in the end find a retailer for any headphones on your shortlist, and if at all possible, go visit them. And have a listen. Because the makers of the headphone may have a different vision for how headphones should sound to how your ears work. Or they might in fact accord with your hearing.

And the only way to know for sure is to check for yourself.


* Let’s be super kind for a moment, and allow that placing a piece of cardboard under one leg of your listening couch actually changes the sound in some perceptible way. What if it is actually making it worse? Why this assumption that a change is for the better?