Improving the Effectiveness of Paired Comparison Tests for Automotive Sound Quality

Sounds generated by new vehicles are important for brand identification and product differentiation. Consequently automotive manufacturers are interested in how to turn customer preferences of sounds into achievable engineering targets; sound quality engineering. One tool within this process is paired comparison testing, where a jury is asked to choose between pairs of sounds given a particular question, e.g. “Which is more powerful ?” or “Which is more refined ?”. Such evaluations are costly and time consuming; hence artificial neural networks (ANNs) are being developing to reduce the reliance on jury evaluations. To evaluate n sounds, n(n-1) pairs are presented to the jury. This includes pairs that are presented to the jury in i-j & j-i orders. When preferences are averaged over jurors, pair probabilities can be calculated e.g. the probability that the sound i is preferred to sound j (pi,j) and vice versa, which should sum to unity. However in practice due to repeatability this is not always the case as juror responses maybe influenced by the order in which the sounds are played, similarly the pair may be difficult to compare. Only one probability can be used to train the ANN. To overcome these difficulties a paired comparison test using freeplay has been developed and assessed. In ‘freeplay’ a juror is presented with a single pair and can play and replay the sounds in any order before making their choice, thus removing the effect of pair order. Additionally the results indicate increases in jurors’ repeatability and consistency measures. 1 W/1/009, Jaguar Engineering Centre. Abbey Road, Whitley, Coventry. CV3 4LF. United Kingdom 2 Sound & Vibration Technology Ltd. Station Lane, Millbrook, Bedfordshire. MK45 2YT. United Kingdom