Sound in Space: towards a unified theory of auditory localization

点击翻译按钮获取中文摘要

One of the major challenges to the auditory system is to track sound sources as they move to predict their future path and guide action (e.g. avoid an approaching car). Yet, dynamic sound localization, our ability to determine the position of sounds as they move in space, has largely been understudied due to the complications of laboratory setups. The long-term goal of this program is to understand the perceptual and cognitive underpinnings of auditory motion in an attempt to reconcile current theories of static and dynamic sound localization. Our recent work has demonstrated the existence of upper limits of auditory motion in a range of velocities that had not been explored before. Based on this work, we hypothesize that auditory motion is based on the comparison of static localization cues at successive points in time, constrained by the temporal resolution of the binaural system (blurred snapshot model). But our work so far focused on azimuthal localization using sounds revolving at a fixed distance on the horizontal plane. To test our hypothesis and the relative contributions of different localization cues, we will extend our investigation to trajectories resulting in changes in intensity, direct to reverberant ratio and Doppler shifts. To do so, we first need to characterize spatial audio rendering techniques in terms of their ability to preserve localization cues at the listeners' ears. We will then systematically investigate the effect on monaural and binaural localization cues on motion perception. In later years, we will examine the effect of spatial features on grouping processes of dynamic sound sources in complex auditory scenes with multiple simultaneous trajectories. In parallel, we will implement models of sound localization informed by the results of experiments with human listeners. We will proceed by a progressive complexification, starting with a simple model of azimuthal localization of sounds at a fixed distance, then adding motion-induced blur, then modeling distance and velocity estimation, and finally accounting for our ability to track multiple sounds sources simultaneously. In Year 5, we will integrate the different cues in a Bayesian observer model for optimal prediction, similar to the ones used for visual motion perception. The predictions will be compared with empirical findings to test alternative theories of sound localization. On theoretical grounds, this research program will provide grounds for a unified theory of sound localization. This program has practical implications for spatial sound reproduction (in virtual reality, entertainment, videoconferencing), for improving spatial functionalities in hearing aids and for acoustic monitoring of target sounds. Future directions include multisensory integration in motion perception by investigating how the visual and auditory system can provide complementary information about the movement of objects.