Multi-dimensional listening test: Selection of sound descriptors and design of the experiment

Abstract

A method for the selection of parameters used in multi-dimensional listening tests is presented. It allows the presentation to subjects, through a very short learning process, of a complete set of parameters which are unambiguously understood. This set is built after a first test in which other listeners are asked to freely describe differences and similarities between sounds. A method of analysing these free verbalizations allows the determination of peculiarities for sounds belonging to the same context. Two multi-dimensional test methods are compared. The method in which all sounds (or pairs of sounds) are evaluated according to each parameter give slightly more reliable results than the classical method, in which each sound or pair of sounds is evaluated according to all parameters before repeating the operation for the next pair or sounds. These two methods are examined using the idling noise of small diesel cars.