Abstract
<p>Human listeners effortlessly understand speech which they hear at a wide range of speaking rates. Yet the neural mechanisms that support the flexibility of successful speech processing remain unclear. Recent evidence suggests time-locked neural responses to features of connected speech (e.g. the amplitude envelope) in auditory brain regions, yet key questions about whether and how these brain mechanisms encode speech at different rates remain. Fixed accounts suggest rate-invariant neural encoding mechanisms, whereas adaptive accounts predict rate-adaptive neural encoding in which speech processing adjusts to changes in speech rate. To adjudicate between these hypotheses, we measured neural responses to speech (podcasts, and sentences) presented at three different rates while neural activity was measured using magnetoencephalography (MEG) or stereoelectroencephalography (sEEG). For both recording modalities we observe hierarchical organisation of neural encoding of the speech envelope across the cortex within the left hemisphere: early auditory regions encode speech in a largely rate-invariant manner, whereas higher-order temporal and frontal lobe regions display rate-adaptive encoding, reflected in both changes to the temporal structure of neural responses and in the ability of neural encoding models to generalize across different speech rate conditions. Together, these findings reveal a general computational principle whereby flexible speech perception across temporal scales is supported by a hierarchical transformation from early rate-invariant to later rate-adaptive neural processes.</p>