Neural and autonomic predictors of future listening errors.

Some of the most demanding perceptual tasks require sustained vigilance to detect and classify infrequent, unpredictable targets within continuous sensory streams. Here, we designed a sustained listening task to ask whether cardinal classes of perceptual errors (misclassifications, miss, and false alarms) were associated with variability in event-related processing, ongoing cortical state, and autonomic arousal. We recorded 64-channel scalp EEG and pupil diameter in 39 adults while they monitored streams of random tone pips for occasional target sequences of repeating tones. A target-evoked negativity varied systematically with listening outcome: it was largest for correct reports, reduced for misclassifications, and nearly absent for misses. Notably, these differences emerged within the first 500 ms of the target sequence, before the information needed to classify its length was available. Related negativities also accompanied false alarms despite the absence of a target. By contrast, neural synchronization to acoustic cues in individual tone pips did not differ across perceptual outcomes. We then looked further back in time, to the 1 s period preceding target onset, and identified additional signatures of forthcoming misclassifications and false alarms in pupil diameter and EEG aperiodic slopes. Models evaluated with leave-one-subject-out cross-validation showed that baseline state measures and event-related negativity amplitude provided complementary information about the probability and type of forthcoming errors. These findings link perceptual errors during sustained listening to higher-order cortical and arousal-state dynamics rather than fluctuant encoding of bottom-up acoustic cues.