Conference on Cognitive Computational Neuroscience

Auditory memory is critical for nearly every daily task we perform involving sounds. However, memory is also imperfect, and the reasons for memory errors remain poorly understood. To investigate this, we probed auditory memory using a sound recognition experiment. Recognition was well above chance but decreased with the number of intervening stimuli between the first and second presentations of repeated sounds. Some sounds were misjudged as novel despite being repeated, while others were falsely recognized as repeated despite being new. To understand these effects in normative terms, we developed an ideal observer model using a statistical representation of auditory texture stimuli. The model encodes experiences as memory traces corrupted by noise that grows with time. The model performs the memory task by deciding whether a stimulus is more likely to have come from the distribution of stimuli implied by its memory traces or from the prior distribution over all textures. The model qualitatively mirrored human performance trends when this prior was estimated from a large set of natural textures, but not if it deviated substantially from this natural distribution. The results suggest that humans have internalized a prior over stimulus statistics in memory and use that to make memory judgments.