Speaker
Description
Recent soundscape research commonly adopts the ISO 12913 circumplex to evaluate and predict perceptions of acoustic environments. This psychometric model represents soundscape perception in a circular space defined by the latent dimensions Pleasantness and Eventfulness, with eight Perceived Affective Quality items (pleasant, annoying, calm, chaotic, vibrant, monotonous, eventful, and uneventful) uniformly distributed around its perimeter. However, the structural validity of this circumplex has rarely been examined within predictor-driven modeling pipelines. This study addresses that gap by estimating PAQ item angles, with uncertainty, from ordinal questionnaire ratings conditioned on psychoacoustic and acoustic predictors. We also quantify the relative contributions of Pleasantness and Eventfulness to each PAQ item in order to identify dominant dimensional loadings. Using Bayesian multivariate hierarchical models, we evaluate the circumplex structure across multiple soundscape datasets while accounting for person- and scene-specific effects. The results show that Eventfulness-related PAQ items deviate systematically from their ISO reference positions, while the circumplex geometry remains largely preserved. Eventfulness appears less clearly separable from Pleasantness-related variation than assumed by the ISO model. This suggests that prediction performance based on dimension scores should be interpreted cautiously, as high accuracy may partly reflect coupling between dimensions rather than clean recovery of independent perceptual axes. Practically, the findings suggest that applications of the ISO 12913 circumplex in prediction pipelines could be refined by explicitly assessing PAQ item positions and circumplex geometry, enabling more robust evaluation of whether predicted dimension scores preserve the intended perceptual structure.