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

Self-motion is an essential but often overlooked component of sound localisation. As the directional information of a source is implicitly contained in head-centred acoustic cues, that acoustic input needs to be continuously combined with sensorimotor information about the head orientation in order to decode to a world-centred frame of reference. When utilised, head movements significantly reduce ambiguities in the directional information provided by the incoming sound. In this work, we model human active sound localisation (considering small head rotations) as an ideal observer. In the evaluation, we compared human performance obtained in a free-field active localisation experiment with the predictions of a Bayesian model. Model noise parameters were set a-priori based on behavioural results from other studies, i.e., without any post-hoc parameter fitting to behavioural results. The model predictions showed a general agreement with actual human performance. However, a spatial analysis revealed that the ideal observer was not able to predict localisation behaviour for each source direction. A more detailed investigation into the effects of various model parameters indicated that uncertainty on head orientation significantly contributed to the observed differences. Yet, the biases and spatial distribution of the human responses remained partially unexplained by the presented ideal observer model, suggesting that human sound localisation is sub-optimal.Author summary: By moving our heads, we can obtain additional information about the direction of a sound. This requires the integration of acoustic and sensorimotor information. To understand this process better, we formulated an ideal observer model for active sound localisation, which provided the Bayesian optimal response, given the available information. We then compared the model’s predictions to the results from a behavioural localisation experiment with sources presented from loudspeakers at a wide distribution of directions. While the model generally matched human performance, it could not accurately predict the bias and spread of localisation estimates for stimuli from certain directions, most notably from above and behind the listener. We found that uncertainty about head position played a significant role in this discrepancy. Still, the distributions of human responses were not fully explained by our model, suggesting that humans may utilise the information available to them sub-optimally.