Loudspeakers: Objective evaluations - Part 3: Comparing the subjective and objective domains

[Part 1 begins with a look at the two basic radiation patterns of sound sources. Part 2 answers the question "If we want to measure a loudspeaker, and from those measurements try to anticipate how it might sound in a room, what should we measure?"]

18.3 COMPARING THE SUBJECTIVE AND OBJECTIVE DOMAINS
Figures 17.2 and 17.3 showed that even primitive measurements and listening tests reveal a subjective preference for loudspeakers with frequency responses that are flat and smooth. These results were not published at the time; they were not really in the category of scientific data, but they provided a stimulus to do more. Interest was not lost, and activity continued, but it was many years before the vagaries of life and work conspired to produce the circumstances for the next big step.

18.3.1 Measurements
It has been said before that measurements don't change. If they are done properly, they can be repeated many times in many places and the answer is closely similar. Opinions are different. Not only opinions from different persons, but from the same person at different times and places. In audio, it has been long regarded as a "given" that personal taste in sound quality is variable, not really to be trusted in any generalized sense.

In 1986, almost 20 years after the amateurish first tests described in Chapter 17, the author performed an elaborate subjective-objective investigation. It led, as discussed in Chapter 17, to a selection process for listeners. Of the 42 listeners who started the tests, only those results from the 28 most consistent are included in the following results. They all had hearing threshold levels within 10 dB of ISO audiometric zero at frequencies below 1 kHz and within 20 dB up to 6 kHz (Toole, 1986, Part 2). All of the listening tests were double-blind, of course; care was taken to avoid biasing the listeners.

Listeners auditioned the loudspeakers in mono in groups of four, presented in different randomized combinations, until each listener heard each loudspeaker three to five times. After completing a questionnaire of subjective qualities, listeners were required to provide an overall "fidelity rating" on a scale of 10, where 10 describes the best imaginable and 0 represents the worst imaginable reproduction.

No reference sound was provided, and there was no formalized training of the listeners. Most had no prior experience in structured listening tests, although all had experience in critical listening either in their professions or audio hobby.

Figure 18.12a shows some results from the tests, sorted into columns according to the overall fidelity rating, and into rows according to the combined vertical and horizontal angular range embraced by the measurements.

FIGURE 18.12 (a) A sample of results, showing loudspeakers grouped according to subjective fidelity ratings in three categories. There were 6 loudspeakers awarded ratings between 7.5 and 7.9, 11 loudspeakers in the range 7.0 to 7.4, and 7 loudspeakers in the range 6.5 to 6.9. The original data include a fourth, lower category. The measurements are unsmoothed, 200-point, log-spaced stepped-tone anechoic measurements. To eliminate the effects of loudspeaker sensitivity, the vertical positions of the curves were normalized to the mean sound level in the 300–3000 Hz band. This same frequency band was used to normalize listening levels. (b) An enlargement of the upper left graph in (a). From Toole, 1986, Figure 7.

It is not difficult to see a progressive degradation in the smoothness of the curves as the fidelity rating decreases. The paper includes a lower category that is even less regular. It is important to note two trends here:

• There is an underlying "flat" trend in these clusters of curves. The variations, even the larger ones, seem to be fluctuations around a horizontal line for the on-axis groups and around quite straight gently-sloping lines for the off-axis groups.

• The average bass extension - the low-cutoff frequency - progressively decreases as the fidelity rating increases. The listeners liked low bass - not more bass, in the sense that it is boosted, but bass extended to lower frequencies.

Figure 18.12b shows an enlarged version of the top-left group of curves, the on-axis measurements of the highest rated loudspeakers. They have been vertically shifted to be symmetrical around 0 dB, and a ±3 dB tolerance band is shown. Clearly a ±3 dB numerical description does not do these loudspeakers justice.

It seems evident that smooth and flat was the design objective for all of these loudspeakers. Deviations from this target are seen at woofer frequencies (below 150 Hz), in the woofer/midrange to tweeter crossover region (1 to 5 kHz), and in the tweeter diaphragm breakup region above 10 kHz. The variations seen among these six loudspeakers from different designers and manufacturers, and even different countries, are smaller than the production tolerances allowed by some manufacturers for a single model. Over 20 years later, these loudspeakers would not be embarrassed if compared to products in today's marketplace.

The apparent preference for extended bass motivated Figure 18.13, where the low-cutoff frequencies for all of the loudspeakers were determined at two levels relative to the reference 300–3000 Hz band.

FIGURE 18.13 Plots of the low cutoff frequencies determined at -5 dB and -10 dB relative to the average sound level over the 300–3000 Hz band.

The normal -3 dB "half-power" level was also tried, but it yielded no relationship, which was anticipated because of the substantial effects of solid angle gains (Chapter 12) and bass resonances (Chapter 13) in the room in which the listening tests took place (or indeed any room). Therefore, it was not a total surprise to find that the relationship between the fidelity rating and the low-frequency cutoff reached a maximum correlation for cutoff frequencies determined at the -10 dB level. Bearing in mind that this is a correlation achieved with all other factors varying indicates that bass extension is a very important factor in overall sound quality evaluations.