Between the Ears: the art and science of measuring headphones

Headphones get pretty short shrift in much of the hi-fi press, which is puzzling—the headphone market is burgeoning. I don't know what the equivalent US figures are, but in recent years the UK headphone market has increased by an annual 15–20% in both units sold and overall revenue. It's easy to dismiss this as a natural byproduct of the Apple iPod phenomenon, but 20% of the market value is now accounted for by headphones costing over $120; a significant subset of consumers would seem to be looking for quality. When you also consider that many people's first exposure to higher-quality audio comes via headphones, there is ample reason for treating them more seriously.

Does this explain why I recently shelled out $3500 to equip myself for headphone measurement? Only partly—there were other justifications as well. If you believe that measurement has a role in designing and assessing audio products, then that philosophy must apply to everything—even products, such as headphones, that are notoriously difficult to measure meaningfully.

Some personal motivation was also involved. My first "hi-fi" system (I would dread to revisit it) was assembled around headphones for a number of very good reasons: I could afford them; I could carry them, just, from London's Tottenham Court Road back to the waiting school train at Charing Cross (along with a BSR McDonald MP60 turntable, afrormosia plinth, Goldring G850 cartridge, and a Trio (Kenwood) amplifier whose model number I've forgotten); and I could listen to the Trio headphones as long and as loud as I liked without censure. Yet the headphone experience palled, and my subsequent headphone listening over the years has often left me wishing for better, even when the headphones concerned were way superior to my original Trios. Somewhere inside, a voice keeps nagging me that headphone performance is, too commonly, not everything it could or should be.

But, as I've hinted, measuring headphones meaningfully is no simple proposition. The first and most fundamental problem is obvious simply by observing your fellow man: We all have differently sized and shaped heads and pinnae (outer ears), and that means we all experience headphones differently—even the insert types that press into the outer end of the ear canal and thereby bypass the pinna. This isn't just a matter of variable acoustical interaction between headphone and ear: it also involves our auditory expectations. We each have a unique set of head-related transfer functions (HRTFs), the term given to the frequency response pertaining at each ear for a flat-spectrum sound source at a given location relative to the head. We use these HRTFs, which change as the location of the sound source changes, to help localize sound, and the brain is able to compensate for these response modifications so that the timbre of the sound source remains unchanged as it moves. Given the significance of our individual HRTFs to our perception of the auditory world, it's no surprise that different people perceive the imaging and tonal characteristics of a given headphone somewhat differently (footnote 1).

In recognition of this biological variability, a favored technique for determining headphone frequency response—as described in the ITU-R BS.708 recommendation, for instance—involves using a tiny probe microphone to measure frequency response near the entrance of the ear canals of a variety of different human subjects. In the case of ITU-R BS.708, it is specified that at least 16 people be used and their results averaged. Researchers and engineers with access to willing students or colleagues can realize this procedure, given enough time, but it is obviously impractical for the headphone reviewer. What we need is an artificial ear of some form that will give results that are consistent, broadly representative, and quick to obtain.

Here we have a variety of options. We could use an acoustic manikin (the proper term for a dummy head), aka a head and torso simulator (HATS), with microphones placed at the outer end of where the ear canals would be. But we have only to scan today's headphone market to appreciate that this option is not the best. It would allow circumaural, supraaural, and earbud headphones to be measured, but not insert types. (I use earbud to describe headphones that fit into the concha, the large depression in the center of the ear that leads to the ear canal.) As increasing numbers of insert headphones are now available to audiophiles, this is a significant disadvantage.

We could opt instead for a manikin incorporating eardrum simulators, which would allow insert 'phones to be measured as well. But unless you have other uses for it, any kind of manikin is an expensive option. Aside from the cost of the artificial head and torso, a manikin incorporates two artificial ears—two artificial pinnae, two eardrum simulators, two microphones—and requires doubled-up electronics as well. Despite which, we will usually measure one headphone channel at a time, particularly if using loudspeaker-measurement software. So the second ear is redundant and another unnecessary expense.

Far easier on the pocket (although not cheap: professional acoustical measurement hardware never is) is a single artificial ear of the type I now have, more formally known as an ear-and-cheek simulator because it also incorporates a surface against which the earpad of a circumaural headphone can rest.

Some artificial ears of yore, you may recall, had nothing resembling a real human pinna. They simulated the gross acoustic impedance of the ear, but not the complex interaction of a circumaural or supraaural headphone with the folds of the outer ear. B&K's 4153 artificial ear, for example, looked more like a complex fuel coupler than an ear. It's still available.

Footnote 1: For evidence of just how large the HRTF disparities are, see H. Møller et al, "Design Criteria for Headphones," Journal of the Audio Engineering Society, Vol.43 No.4 (April 1995).