Accurate sound localization is an important ability: it informs listeners about their environment without the need to point their eyes at everything, and it serves to guide the eyes towards relevant events; it even works in darkness or when direct eyesight is obstructed. However, when human sound localization is tested in the laboratory, with many sound speakers in a semicircle from left to right in front of the listeners, systematic errors (i.e., biases) are often observed. For example, individuals regularly localize sound sources consistently nearer to- or further away from the frontal midline than the true speaker locations. Confusingly, such biases sometimes point in opposing directions for different participants and experimental conditions. Moreover, it seems that the locations of preceding sounds affect the perceived location of the current sound, which can be both attracted to- and repulsed away from the preceding sound locations, depending on the precise circumstances.
While such sound localization biases have often been regarded as mere imperfections of the auditory system, this project views them as epiphenomena of two fundamental mechanisms that normally aid directional hearing in uncertain and dynamic auditory environments. First, the sensory system adapts to improve spatial precision in the range where it recently experienced sounds, thus leading to local contrast enhancements but also unintended spatial distortions. Second, perceptual stability over time is achieved by integrating auditory signals with existing beliefs about sound sources, but this may cause cognitive expectations to bias spatial estimates. The consequences of these two processes interact, and this can potentially explain the complex and opposing patterns of response biases that are observed in sound localization studies.
Building on recent research findings from vision science, various implementations of the two processes will be simulated in a multitude of computer models with individualized parameter values, with the aim of comparing the models’ sound localization predictions to the observed localization responses of human participants in carefully conducted laboratory experiments with diverse environmental manipulations. Through the selection of computer models whose systematic error patterns best resemble those of human listeners we will learn about the brain’s computational mechanisms that underlie generally accurate dynamical spatial hearing but produce noticeable biases under particular conditions. The project thus enhances our fundamental understanding about when and why systematic errors in sound localization occur, and so it intends to reconcile and explain a growing body of apparent contradictory research findings that portray biases in opposing directions.
The BIASED EAR project receives funding from the Austrian Science Fund (FWF).
Grant-DOI: 10.55776/PAT1267925
