By Julian D. Hirsch
SPEAKER TESTING AGAIN: Subsequent to examining in last month's column the testing philosophy behind Consumers Union's recently published speaker ratings, I received the following communication from Roy Allison, President of Allison Acoustics, Inc. He out lines, in somewhat greater detail than I did, how CU arrives at its numerical ac curacy ratings, and he includes his perception of some of the flaws in their procedures as well.
----------------
CONSUMERS UNION'S latest report on loudspeakers (Consumer Reports, February, 1976) demonstrates a problem common to all equipment reviewers: that of making valid measurements of loudspeaker performance.
The problem is uniquely difficult and controversial for loudspeakers among the hi-fi components, because the output signal to be measured is not electric but acoustic, and it is sent outward into three-dimensional space.
Sound-pressure output vs. frequency varies not only with the direction of radiation but also with time. The acoustic energy produced is strongly influenced, over a wide range of frequencies, by the loudspeaker's working environment (the listening or test room). Finally, after modifying the energy output, the room redistributes this energy in resonance-mode patterns. Without question, it is a complex situation.
It is not surprising, therefore, that the particular aspects of a loudspeaker's performance which correlate best with what we hear are a matter of dispute. CU's test engineers believe that the most important characteristic of a loudspeaker is its total acoustic-power output vs. frequency. Many share that opinion (as do I). Where, then, did CU go wrong? The answer lies partly in measurement technique and partly in an oversimplified criterion of accuracy.
CU uses an anechoic chamber to measure the total acoustic power produced by a loud speaker. Such a test chamber has walls that absorb sound energy almost completely. With no reflections from the room boundaries, there can be no resonance modes and the test microphone is not confused by room effects; it responds only to the direct radiation from the loudspeaker. The measurement procedure is thereby simplified. Unfortunately, the room's influence on the loudspeaker's power output (which can be as much as 20 dB) is also eliminated. Here is a classic case of throwing out the baby with the bath.
Using a sophisticated computer-aided procedure [described in this column last month], CU obtains a very accurate measurement of power output vs. frequency in an anechoic environment and proceeds to analyze the data by'/-octave frequency bands within the range from 110 to 14,000 Hz. The computer adjusts the overall loudness level to an arbitrary but reasonable value, then assigns the appropriate sone value (a loudness unit) to the relative output of the loudspeaker in each'/- octave band. For each band it divides the sone value of the test loudspeaker by the sone value a theoretically perfect loudspeaker (one having perfectly flat power response) would have in that same band if played at the same overall loudness level. If the test loudspeaker matches the ideal in a given 1/2-octave band, the quotient of this division process is 1.00; if it is 1 dB up or down in level, the quotient is approximately 1.08 or 0.92; and if it is off by 9 or 10 dB in that band, the quotient turns out to be about 2.00 or 0.5, signifying double or half loudness.
Computing the "accuracy score" is then a simple process. The absolute amounts of the deviations from perfect-match quotients of 1.00 are averaged for all the twenty-one bands, the result is subtracted from 1.00, and the remainder is expressed as a percentage.
Thus, if the average deviation from flatness of power response were 1 dB, the average loud ness difference from the ideal would be 0.08 sone, and the accuracy score would be 1.00-0.08 = 0.92 x 100 = 92 percent. The fact that an average difference of 1 dB represents eight percentage points also explains CU's caveat about an eight-point ambiguity in the ratings:
in theory, the average listener can just detect 1-dB differences.
This system has at least two grave flaws.
The first has to do with the unqualified premise that a perfectly flat power response is necessary for accuracy. STEREO REVIEW has published discussions of this point in the past, and I will not pursue it now. But let us postulate two loudspeakers, A and B, with quite different characteristics. Speaker A has a very smooth power-response curve, straight but tilted down at the high-frequency end, with the tilt easily corrected (if desired) by means of a treble tone control. Speaker B has a power-response curve that averages flat from 110 to 14,000 Hz, but it has a big hole in the output at 1,000 Hz and a nasty peak at 3,000 Hz. There is no doubt as to which will be the better-sounding loudspeaker, but, in the CU computer's eyes, they may well be equal because the average deviation in loud ness from the "ideal" is the same.
The other major problem is at the low-frequency end of the range. CU claims "very high" correlation between their accuracy score ratings and listening tests in their music room, provided the program material fed to both the reference and test speakers is limited to the range above 110 Hz. They do not include any data below 110 Hz in the accuracy score because, they say, they cannot predict the effect of a listening room on the bass out put of a loudspeaker.
What CU does not point out is that this very high correlation is obtained with the speakers positioned at least 31/2 to 4 feet away from any surface of their listening room, which puts the adverse effects of boundary reflections nicely below 110 Hz. But I have yet to see a living room setup with loudspeakers that far away from the floor and walls, and I doubt that many exist. A more practical spacing is 1 to 11/2 feet from woofer to wall. If CU were to put the loudspeakers within this more realistic distance from their listening-room wall, their anechoic accuracy scores would not correlate with listening tests below about 400 Hz. In terms of representing what loudspeakers will deliver to real rooms in practical circum stances, anechoic tests are invalid below that frequency, and so are CU's accuracy scores.
In case you think that may not be important, consider: 40 percent of the audible frequency range lies below 400 Hz.
CU-and other organizations in the business of reporting on loudspeakers, as well really should abandon anechoic chambers, at least for tests below the middle range, and assess loudspeaker performance in an environment more nearly representing reality.
That is to say, the test room should contain at least three intersecting hard surfaces and the loudspeaker should be placed in a typical position with respect to these surfaces. Then a sufficient number of response curves must be taken in the far field of the room so that, when integrated, the power output is obtained for a practical operating condition of the loud speaker. Only in that way can a valid low-frequency measurement be made.
- Roy F. Allison
MR. ALLISON'S point about the conditions necessary for close correlation between the numerical accuracy rating and the subjective rating accorded to a speaker by a listening panel is well taken. This is also one of the major limitations of the simulated "live-vs.-record ed" test that I perform, which is useful only above 200 Hz. It is possible (though rare) for a test speaker to be literally 100 percent accurate in this test, yet have a plainly audible and even objectionable lower-mid-range or bass coloration that makes it a far from "perfect" speaker when used to play wide-range program material.
In regard to Mr. Allison's final suggestion, I would agree in principle that anechoic chambers are of little value for establishing the "accuracy" of a speaker to be used in a normal home environment. In fact, at middle and high frequencies, my own test procedure is not very different from that proposed by Mr. Allison. Fortunately, a simple technique exists for measuring the bass output of a speaker independent of room characteristics. A paper presented by D. B. Keele of Electro-Voice to the 45th Convention of the Audio Engineering Society shows that it is possible to obtain the equivalent far-field response of a woofer by a near-field pressure measurement. The test room plays no part in the measurement, so that it neither degrades nor enhances the bass-response measurement.
Mr. Allison's suggestion that low-frequency response be determined by multiple measurements in a "live" room is, I feel, quite impractical, aside from the considerable time necessary to make such measurements. Although some speakers, such as the Allison models, are designed to make use of room boundaries to deliver a flat bass response, and thus would appear to best advantage when tested in that manner, others probably would not. It could reasonably be claimed that another room might give very different results. This is equally true of the Keele method, or any other except a measurement in your own room. Let us be careful not to confuse (1) the actual performance of a speaker with (2) our measurement of that performance, for they are two very different things! However, the Keele test, which we use, does show a capability or potential performance of the speaker, which is of course subject to modification by its operating environment.
In conclusion, I would like to remind the reader of an unfortunate fact of hi-fi life: the only way to find out what a speaker sounds like in your own home is to set it up there and listen to it. Sorry there are no short cuts!
Tested This Month
Akai GXC-325D Cassette Deck Stax SR-5 Stereo Headphones Design Acoustics D-2 Speaker Lenco L-85 IC Turntable
Also see:
GOING ON RECORD, JAMES GOODFRIEND
THE BASIC REPERTOIRE, MARTIN BOOKSPAN