Natural ITD statistics are represented in the neural code underlying human sound localization

Neural representations of the statistical structure of sensory cues may optimize perception. We investigated this hypothesis in human sound localization, focusing on the statistics of interaural time difference (ITD), a major cue for inferring azimuth location of sounds. Natural ITD rate of change across azimuth and a novel metric of ITD variability over time were estimated from human head-related transfer functions (HRTFs) and modeled cochlear filters to obtain a Fisher information metric of ITD. Using protocols which disabled ITDrc estimation and delivered zero-ITDv stimuli, we found that spatial discriminability and novelty detection are consistent with the hypothesis that the brain contains an adapted code that represents natural ITD statistics, exploiting them for optimizing discrimination. We further show that natural ITD statistics could be represented under coding frameworks postulated by classic neural models, highlighting the validity of these models and providing potential explanation for the observed distribution of best ITDs across frequency found in the mammalian brainstem. These findings support the hypothesis that natural sensory statistics are represented in neural networks underlying perception.