ARL-TR-8672 - Characterizing Asymmetrical Ratings of Similarity for Real-World Complex Environmental Sounds | U.S. Army Research Laboratory

摘要

Understanding complex environmental sound perception is critical for understanding human behavior in real-world settings. Whereas simple stimuli can be processed on the basis of physical characteristics alone, environmental sounds contain semantic and contextual information, which can lead to asymmetrical similarity ratings. The goals of this study were to 1) quantify asymmetries in pairwise similarity ratings of 25 environmental sounds, 2) test the hypothesis that these asymmetries are systematic order effects, and 3) characterize the impact of asymmetries on similarity spaces constructed using multidimensional scaling (MDS). First, 26 participants rated the similarity of every pair of the sounds (from 1 = most similar, to 7 = as different as possible) in both orders. In the second experiment, participants were asked whether they could identify the source of each sound. The relative identifiability of each sound in the pair influenced rated similarity; presenting the more identifiable sound in the base position (second) produced higher-rated similarity. MDS spaces constructed using randomly assigned presentation orders were highly intercorrelated, whereas MDS spaces generated from pairs ordered by identifiability (i.e., higher to lower vs. lower to higher identifiability) shared roughly 10% less variance, indicating that randomization is effective for controlling for order effects in pairwise similarity ratings.