Previous exposure to sensory feedback noise causes a decrease in online compensation for sensory perturbations in speech

Ben Parrell* & Carrie Niziolek*

University of Wisconsin–Madison, Department of Communication Sciences and Disorder

*equal contribution

In motor control, sensory afferent information is incorporated with forward model predictions into a single estimate of the current state of the motor plant, which is then used by the controller to generate motor commands. This is often modeled as a (quasi-)Bayesian process, where the weight assigned to each signal is proportional to its reliability. Consistent with this hypothesis, previous work in a reaching task has shown that increased noise (decreased reliability) in the visual signal leads to a decrease in the use of visual feedback for hand position estimation (Kording & Wolpert 2004). However, how long-term exposure to unreliable feedback affects the sensory integration process remains unclear. Here, we leverage the partial compensation for auditory perturbations seen in the speech motor control system to test the hypothesis that persistent exposure to unpredictable noise in sensory feedback changes the relative weighting of sensory and predictive signals. In a within-subjects experiment, we first manipulated the reliability of the auditory feedback received during speaking (exposure phase), then assessed the compensatory response to large auditory perturbations (test phase). In separate sessions, participants were exposed to veridical feedback or small feedback perturbations (standard deviation, 15 mels) in the exposure phase. We test two competing hypotheses. H1: If the feedback mismatch during the perturbed exposure phase is attributed to inaccuracies in forward model prediction, reliance on sensory feedback should increase, leading to a larger compensatory response to test-phase feedback perturbations. H2: If the mismatch is attributed to errors in the sensory system, reliance on sensory feedback should decrease, leading to a smaller compensatory response in the test phase. Results show that participants produce smaller responses to auditory perturbations during the test block when following the perturbed exposure phase compared to the veridical exposure phase. This smaller response, consistent with H2, suggests that exposure to random sensory prediction errors causes a downweighting of sensory feedback and increased reliance on forward predictions.