TESTING, QUALITY & STANDARDS: Measuring Sound Quality
Sound quality assessment is about how to measure and assess the sound produced by a product. It is possible to use juries of customers to listen to and judge the sound produced, but because everyone is an individual, it is necessary to test many people to get some form of average response. So, jury testing is a rather slow and tedious method. For this reason, people have developed sound quality metrics which are meant to directly relate to human subjective response and can be measured quickly on specialist instrumentation.
For instance, there is a metric called loudness, which (unsurprisingly) relates to a person's perception of the loudness of a product. The loudness metric is more effective at modelling the human response than the better known sound pressure level measured in decibels, as it is based on a more complex model of human hearing. However, the sound quality community is split as to the usefulness of many of the sound quality metrics. Some use the metrics widely in product design, while others claim that they are useless and rely solely on jury testing.
There is a wide range of tools available advertising the ability to quantify sound quality objectively. Companies who sell sound quality hardware and software include: 01dB (dBFA32), B&K (PULSE analyzer), and HEAD acoustics (Artemis and SQ lab II). These systems calculate a bewildering array of sound quality metrics. For each product type, however, only a small number of these metrics will be useful. There is one metric, though that everyone would agree is useful, and works for most products, and that is loudness.
For instance, although washing machine A is the judged to be quietest, it actually generates the most low frequency sound, but it does this at frequencies where the human ear is less sensitive. So, one possible approach to making a washing machine sound better is to shift the dominant noise to lower frequencies. This tactic can work with most noise sources; and motors with more low frequencies usually sound as though they are of higher quality. You don't often hear a person talk about there being a lovely, high-pitched screech coming from their expensive appliance.
The loudness calculation method uses the analysis bandwidths that the ear uses. These are known as the critical bandwidths and are a measure of the frequency resolution of the ear. For example, two tones less than one critical bandwidth apart will not be heard as two separate sounds. Instead, the sounds will partially mask each other, making the perceived loudness not a simple energy summation. This complex masking of one sound by another must be correctly modelled, and there is a published standard for the calculation method.
The loudness of a product usually dominates how we perceive a sound. So the loudness metric also correlates strongly with how pleasant people perceive a product's sound to be, how robust the product sounds and whether the product is perceived as being high quality. In the case of washing machines, the quieter the product the better; however, that isn't true of all products. A manufacturer told us of a case where reducing the loudness of a product (a leaf blower) had caused customers to return the product because they believed that the product was less powerful and, therefore, less useful.
Consequently, for some products, like leaf blowers, it might not just be a case of reducing noise, but of using a bit of auditory deception, producing a quieter product which still sounds powerful. Would shifting the noise to lower frequencies produce a powerful sounding machine? This more subtle approach is already being used in the automobile industry, and, to gauge these finer aspects of the sound, more metrics are needed.
A key differentiator of a noise source is the balance between the high (treble) and low (bass) frequency content. For this, there are two common metrics. Sharpness is a measure of the high frequency content of a sound - the greater the proportion of high frequencies, the sharper the sound. Booming is a measure of the low frequency content of a sound - the greater the proportion of low frequencies, the greater the booming sound. So booming can be considered to be the opposite of the sensation of sharpness.
The loudness metric is a simple sum of the specific loudness values. The sharpness and booming metrics use weighting curves to bias the sum to take more account of high or low frequencies respectively. But the experience of many is that these further metrics don't often correlate with the judgements made by juries.
For instance, researchers at the Acoustics Center tested three different products: kettles, washing machines and leaf blowers, and only found infrequent correlation with the sharpness metric. This happens because the simple weighting approach is only a crude approximation to the complex way that we perceive sound.
Most commercial sound quality assessment software produces a large number of metrics, because it is known that many won't work for a particular product. Many of these metrics are simply variations of more common definitions, in the hope that these will correlate better. It is then up to the user to pick out and use the metrics that do work. One problem with this approach is that these metrics are not standardized, and so it is impossible to compare the output from different sound quality assessment software.
So we have solved the problem of poor correlation by standard metrics, but this is not entirely satisfactory. The customized metrics are highly unlikely to generalize to different products, so we have to derive new metrics for each product of interest. Even more worrying, if someone produces a revolutionary new product type that generates a unique new sound, then these metrics will probably then fail.
Furthermore, these customized metrics are not standardized, and so it is impossible to use them outside product design. You could not, for example, use them on consumer labels because manufacturers would never agree which customized metric to use. This problem has led to a proliferation of metrics being developed. Most of these metrics are commercially confidential, and so the details are not freely available. Having said this, very specialized and tailored metrics are the norm in automobile sound quality, and it is likely that, until much more sophisticated auditory and perception models have been developed, sound quality metrics are going to have to be tailored for individual products for the foreseeable future.
More information about sound quality metrics can be found by visiting www.acoustics.salford.ac.uk/sqa. The Web site allows visitors to listen to the different sounds, and read the best practice guides for objective and jury testing.