Speaker
Description
Spatial hearing depends on internal representations of acoustic cues that vary across space. Elevation perception relies on learned spectral cues from pinna filtering, making it uniquely dependent on prior knowledge rather than instantaneous acoustic information alone. It remains unclear whether listeners track statistical properties of elevation consistent with Bayesian inference and whether this differs across regions of auditory space.To test this, listeners monitored sequences of noise bursts with elevations sampled from a distribution defined by a mean and variance. Each sequence either remained constant or changed in mean and/or variance (mean shifts: 30° or 60°; standard deviations: 15° or 25°). Participants reported when they detected a change.Preliminary results (N = 6) show reliable detection of changes in elevation distributions, with sensitivity increasing for larger mean shifts (d' = 1.4 vs 2.2 for 30° vs 60°, p < .001). Detection was more accurate for sequences with initial distributions in the front than for those starting above (p = .025) or behind (p = .034). This spatial asymmetry may reflect stronger or more reliable prior representations for frontal elevations. The influence of variance is still under investigation, with data collection in progress. Ongoing work will fit behaviour with Bayesian change-detection models to test consistency with Bayesian inference and estimate how latent variables, such as temporal integration, vary across space.