Robust Cortical Encoding of Slow Temporal Modulations of Speech

Abstract

This study investigates the neural representation of speech in complex listening environments. Subjects listened to a narrated story, masked by either another speech stream or by stationary noise. Neural recordings were made using magnetoencephalography (MEG), which can measure cortical activity synchronized to the temporal envelope of speech. When two speech streams are presented simultaneously, cortical activity is predominantly synchronized to the speech stream the listener attends to, even if the unattended, competing-speech stream is more intense (up to 8 dB). When speech is presented together with spectrally matched stationary noise, cortical activity remains precisely synchronized to the temporal envelope of speech until the noise is 9 dB more intense. Critically, the precision of the neural synchronization to speech predicts subjectively rated speech intelligibility in noise. Further analysis reveals that it is longer-latency (∼100 ms) neural responses, but not shorter-latency (∼50 ms) neural responses, that show selectivity to the attended speech and invariance to background noise. This indicates a processing transition, from encoding the acoustic scene to encoding the behaviorally important auditory object, in auditory cortex. In sum, it is demonstrated that neural synchronization to the speech envelope is robust to acoustic interference, whether speech or noise, and therefore provides a strong candidate for the neural basis of acoustic-background invariant speech recognition.