Paired comparison listening tests and circular error rates

On the basis of noises recorded in trains, this article presents circular errors rates obtained by two possible paired comparison methods. The first one allows listeners to answer that two noises in a pair are equally annoying, while the second one forces them to select one of the sounds. It is shown that the latter can lead to very high number of circular errors when perceptual differences between noises are small. Therefore the former method is recommended, and the number of circular errors must be essentially considered as an indicator of the difficulty of the task.