Intelligent systems approaches to product sound quality evaluations - A review

As a product market becomes more competitive, consumers become more discriminating in the way in which they differentiate between engineered products. The consumer often makes a purchasing decision based on the sound emitted from the product during operation, by using the sound to judge quality or annoyance. Therefore, in recent years, many sound quality analysis tools have been developed to evaluate the consumer preference as it relates to a product sound and to quantify this preference based on objective measurements. This understanding can be used to direct a product design process in order to help differentiate the product from competitive products or to establish an impression on consumers regarding a product's quality or robustness. The sound quality process is typically a statistical tool that is used to model subjective preference, or merit score, based on objective measurements, or metrics. In this way, new product developments can be evaluated in an objective manner without the laborious process of gathering a sample population of consumers for subjective studies each time. The most common model used today is the Multiple Linear Regression (MLR), although recently non-linear Artificial Neural Network (ANN) approaches are gaining popularity. To shed further light into these approaches, this paper is written to review the sound quality process and neural network models, and extend these introductions into a discussion regarding the value that can be gained in using an intelligent systems approach, namely ANNs, to sound quality analysis. The paper goes into specific shortcomings that are associated with both the current regression and neural network approaches, and concludes with new thoughts regarding a robust approach to improving the current state-of-the-art technology.