Bayesian active sound localisation: To what extent do humans perform like an ideal-observer?

Area of Science:

• Auditory Neuroscience
• Human Perception
• Computational Auditory Scene Analysis

Background:

• Sound localization relies on head-centered acoustic cues.
• Sensorimotor information about head orientation is crucial for world-centered sound perception.
• Head movements reduce directional ambiguity in sound localization.

Purpose of the Study:

• To model human active sound localization using an ideal observer approach.
• To compare Bayesian model predictions with human performance in a free-field experiment.
• To investigate the impact of head orientation uncertainty on sound localization accuracy.

Main Methods:

• Developed a Bayesian ideal observer model for active sound localization.
• Incorporated sensorimotor information of head orientation.
• Conducted a free-field active localization experiment with human participants.
• Compared model predictions with empirical human performance data.

Main Results:

• Model predictions generally agreed with human performance.
• Spatial analysis revealed discrepancies between the ideal observer and human behavior for specific source directions.
• Uncertainty in head orientation significantly contributed to observed differences.
• Human responses exhibited biases and spatial distributions not fully explained by the ideal observer model.

Conclusions:

• Human sound localization, while generally effective, appears to be sub-optimal.
• Head orientation uncertainty is a key factor influencing localization accuracy.
• The ideal observer model provides insights but does not fully capture the complexities of human active sound localization.