April 1, 2005

Vanished Audio Tests

A curious thing has happened in audio reviewing. On the one hand, what was once a fairly specialized audio component -- the headphone -- has become virtually ubiquitous. It’s still probably fairly rare that most of us would listen to music on our primary system using headphones, except perhaps late at night, but the proliferation of MP3 players and other portable devices means that a huge amount of listening is done with these convenient gadgets clamped on our heads or stuck in our ears. Yet no one seems able to evaluate them, either at the manufacturing or consumer level. As a result, makers of ‘phones and their ultimate users both tend to take an entirely subjective approach to figuring out what’s good and what’s not.

However, there is some science that can take some of the guesswork out of measuring and reporting on headphone sound, and it proved satisfyingly reliable. Pity no one’s used it since a series of landmark tests in AudioScene Canada more than 20 years ago. To my knowledge, no one else has ever published anything like these reviews, which means that a very valuable tool has been lost.

Testing headphones is not as straightforward as testing most other audio components, where flat frequency response is the ideal -- generally speaking, the closer most components come to achieving a flat response, the better they sound. But the very fact that ‘phones interfere physically with the outer ears means that the designer must build into a headphone design some compensation for the bypassed aural functions. A signal that is flat when it reaches the outer ear is anything but by the time it gets to the eardrum, as there is considerable acoustic modification by the ear itself. A headphone alters this by deforming the ear and coupling very closely with the air in the ear canal -- even more so in the case of earbuds, which are placed inside the canal. For a headphone to sound natural, it must exhibit a frequency-response curve that duplicates the modifications the ear would make. Unfortunately, every human ear is different, so no single compensation curve will be ideal.

It is possible, however, to build up a family of curves that represent the range of responses required for the average person to hear a natural sound. These curves, first produced by Dr. Edgar Shaw at the National Research Council in Ottawa, Canada -- and adapted by the redoubtable Dr. Floyd Toole for use in magazine reviews -- are shown as the shaded area on the graphs below. A number of tests performed during the 1970s and ’80s showed that the headphones whose curves most closely conformed to these shaded areas were those that listening panels judged to sound best.

The frequency-response measurements shown were made using a lifelike rubber replica of a human ear of median size, fitted with a microphone at the entrance to the ear canal. The replica was mounted to a form that simulated the side of a human head, and a support some 7" away held the other side of the headphone and allowed the headphone to exert its normal force. The lower curve in each case was measured in this way; the upper curve represents a similar setup, but with some pressure applied to the outside of the earcup.

The shaded area in each of the frequency-response curves indicates what the average human ear "wants" to hear. The measured curves should not be judged by their flatness, therefore, but by their relationships to the range of acceptable responses.

A panel of listeners using a variety of music evaluated each headphone subjectively, without reference to the measurements. There was a high degree of agreement among members of the panel as to the sounds of the various ‘phones, and the subjective results correlated very closely to measured performance: The headphones whose frequency response closely approximated the shape of the shaded area on the graphs tended to sound good to the panel, while the ones whose curves strayed outside the shaded area tended to have problems.

These measurements were made in 1979, and none of the ‘phones tested are available today, which is why I’ve identified them here only by number. But these results are typical of what we might find today, at least to my aging ears, so it should be instructive to look at the range of measured results and see how they relate to the "ideal" range of curves. To get some idea of those relationships, here are some of the comments included in the original test reports.

Headphone 1: In the frequency-response curve, the 'phone matches the "ideal" shaded area well in the upper bass and midrange up to about 2kHz, and again above 10kHz. The gradual rolloff below 200Hz is fairly typical of headphones; and, in fact, it was noted on occasion that the 'phone might have a touch too much bass. Otherwise, our listening panel’s consensus was that this unit had a smooth overall sound, well suited to most of the music we auditioned. The one area where these ‘phones fell somewhat short was in the important 3 to 8kHz range. On the frequency-response curve, it is quite apparent that where the "ideal" curve exhibits a hump, this unit shows a dip. This would correspond to a "hole in the middle" in other components, and it showed up subjectively in our listening tests, particularly on vocal music. A good example of this was during the children’s chorus from Carmen -- an excellent midrange test. While the children seemed farther away than usual, the strength from 10kHz up put their sibilants right up close -- a most curious effect.

Headphone 1

Headphone 2: The sound produced by this 'phone is distinctive. "Dramatic" is a word that might be applied -- one member of our listening panel thought the experience was like having sound "pumped into your head." This, plus the relatively high isolation from outside sounds, makes the sensation of listening with these ‘phones quite unlike that of listening to speakers, or even many other headphones on the market. Part of the reason is the extremely well-maintained low-frequency response, as can be seen in the curve. We are used to seeing a considerable bass rolloff in ‘phones, and that certainly doesn’t happen in this case. At the other end of the spectrum, there are some large peaks in the response -- notably at 3, 6, and 10kHz -- and these tended to give the ‘phones a rather "forward" sound, not unlike speakers that purport to enhance rock music. The wide excursions that can be seen in the curve at the upper end added a bit of coloration to the sound, more apparent on some types of music than others. Popular music tended to fare rather better than classical.

Headphone 2

Headphone 3: The frequency-response curve indicates that the "ideal" curve is matched well by this unit for the most part. The low-end rolloff is to be expected; but it isn’t as severe as it could be, given the lack of a seal between the earcup and the head. Up at the top there is some spectral imbalance, and this was audible, although not particularly serious. From 1.5kHz to 5kHz, the curve is somewhat lower than what we would expect, and this dip was heard as a slightly "distant" effect on some types of music. More important, however, was the fact that the mild weakness in the midrange emphasized the prominence that can be seen in the curve from about 7kHz upwards. Things like record surface noise and tape hiss were given some undue prominence because of this, and certain records -- particularly over-equalized rock records -- could sound a trifle "tinselly." In general, the ‘phones have a smooth natural sound to offer, and a fine sense of space.

Headphone 3

Headphone 4: A look at the frequency-response curve shows that the 'phone approximates the shaded "ideal" area quite closely. Up to 2kHz, and above 6kHz, the match is admirable; and our listening panel found the overall sound very smooth and natural. The dip between 10 and 15kHz lends a slightly muted quality on some program material, but this is far from serious. In the upper midrange -- between 2 and 6kHz -- there is a depression in the curve where the "ideal" 'phone should show a peak, and this was noticed by the listening panel on several types of music. Specifically, choral music seems a bit distant, and some things tend to sound a bit "boxy" -- not uncommon with midrange depressions.

Headphone 4

Headphone 5: The frequency-response curve shows good agreement between the measured response of the 'phone and our nominally "ideal" curve. The low end is very well maintained, possibly because of the good 'phone-to-ear coupling in this circumaural design. In the midrange, there is a broad general depression between 1 and 4kHz, which showed up in the comments of our listening panel. Because it is relatively broad, unlike the sharper dips seen in some other ‘phones, it did not disturb the overall impression of good spectral balance. Above that point, there is a sharp spike at 7kHz, which contributed some coloration in the upper midrange, but the panel found this acceptable.

200504_phone5.jpg (39236 bytes)
Headphone 5

These curves and comments represent a small, more or less random sample of headphones tested on a particular occasion many years ago, but the technique proved its worth over and over. When it was devised, we hoped that manufacturers and other reviewers might follow the lead, but to my knowledge none ever did.

Which is why headphones are still designed and bought "by guess and by God."

...Ian G. Masters


All contents copyright Schneider Publishing, Inc.; all rights reserved.
Any reproduction, without permission, is prohibited.
SoundStage! is part of Schneider Publishing, Inc. and the SoundStage! Network