Context sensitivity across multiple time scales with a flexible frequency bandwidth

Abstract

Everyday auditory streams are complex, including spectro-temporal content that varies at multiple time scales. Using EEG, we investigate the sensitivity of human auditory cortex to the content of past stimulation in unattended sequences of equiprobable tones. In 3 experiments including 82 participants overall, we found that neural responses measured at different latencies after stimulus onset were sensitive to frequency intervals computed over distinct time scales. Importantly, early responses were sensitive to a longer history of stimulation than later responses. To account for these results, we tested a model consisting of neural populations with frequency-specific but broad tuning that undergo adaptation with exponential recovery. We found that the coexistence of neural populations with distinct recovery rates can explain our results. Furthermore, the adaptation bandwidth of these populations depended on spectral context - it was wider when the stimulation sequence had a wider frequency range. Our results provide electrophysiological evidence as well as a possible mechanistic explanation for dynamic and multi-scale context-dependent auditory processing in the human cortex. SIGNIFICANCE STATEMENTIt has become clear that brain processing of sensory stimuli depends on their temporal context, but context can be construed at time scales from the recent millisecond to life-long. How do different contextual time scales affect sensory processing? We show that auditory context is integrated across at least two separate time scales, and that at both of these time scales responses dynamically adapt to a varying frequency stimulation range. Using computational modeling, we develop a rigorous methodology to estimate the time and frequency scales of context integration for separate response components. Our robust results replicated across 3 EEG experiments, and contribute to the understanding of neural mechanisms supporting complex and dynamic context integration.