Temporal context invariance reveals neural processing timescales in auditory cortex

Natural sounds like speech and music are structured at many timescales, but it remains unclear how these diverse timescales are cortically represented. Do processing timescales increase along the putative cortical hierarchy? What timescales are used to code speech and music? Is there hemispheric or anatomical specialization for processing particular timescales? Answering these questions has been challenging because there is no general method for estimating integration periods: the time window within which stimulus features alter the neural response. In this talk, I will describe a simple experimental paradigm (the “temporal context invariance” paradigm) for inferring the integration period of any sensory response. We present sequences of natural sound segments in which the same segment occurs in two different contexts (different surrounding segments), and test how long the segments need to be for the response to become context invariant. By applying this paradigm to intracranial recordings from epilepsy patients (broadband gamma power), we map neural processing throughout human auditory cortex. This map reveals a four-fold increase in timescales between primary (~100 ms) and non-primary regions (~400 ms). Using a separate dataset of responses to a diverse set of natural sounds, we then test what information can be decoded from populations of electrodes with different integration periods. We find that spectral information is best decoded from short integration period electrodes (<200 ms) while sound categories (speech & music) are best decoded at longer timescales (>200 ms). Time permitting, I will also present on-going experiments testing the extent to which neural timescales are fixed vs. adaptive.