Ediz Sohoglu

The paper investigates how prediction error computations influence speech perception in naturalistic listening conditions using EEG and linear encoding models.

The paper shows that neural responses to speech are suppressed by strong predictions when they match input, but enhanced when they mismatch, indicating prediction error computations in auditory regions.

This study uses MEG data to test two Bayesian models of spoken word recognition, finding support for predictive-selection theories over competitive-selection theories.

The paper distinguishes evoked responses from true neural oscillations by showing sustained oscillatory responses after rhythmic stimulation in speech perception.

The paper shows that neural responses to degraded speech show prediction error computations, with signal quality and prior expectations interacting to shape representations within 250 ms.

The paper uses fMRI and multivariate pattern analysis to show that acoustic features are represented independently in primary auditory cortex and integrated in parietal cortex.