Understanding how customers make their decisions on product sound quality

摘要

A product's sound can reinforce its brand image. Sound quality engineering is concerned with turning customer preferences, brand attributes and marketing information into product engineering targets. Automotive companies for example are interested in how their new car should sound, since sound is an important way to differentiate products. Currently, candidate sounds are evaluated by costly and time consuming jury evaluations, often using the method of paired comparisons. Neural networks have been identified as a method of evaluating a large number of sounds quickly and cheaply. A Multi-Layer Perceptron has been trained from existing jury data to predict the probability of a juror selecting one sound of a pair. These results can simulate a jury evaluation and give rankings of new and existing sounds. Neural network inputs include objective sound metrics - numeric or categorical inputs that describe the sounds. However, the difficulty is to decide which of the numerous, time-varying metrics to use and how to represent them as single numbers. Further evidence is required to choose the optimum inputs to present to the neural network. One option is to choose the inputs based on an understanding of how customers actually make their decisions. To this end, a simple interview procedure has been developed and tested. This is described and sample results are presented and assessed. The suitability and wider application of this procedure is discussed, and further work suggested.