Linear modeling of brain activity during selective attention to continuous speech: the critical role of the N1 effect in event-related potentials to acoustic edges

Abstract

Recent work in the field of neural speech tracking provided evidence for a cortical representation of speech through superposition of event-related responses to acoustic edges, an idea closely related to the popular linear modeling approach to study cortical synchronization to speech via magneto- or electroencephalography (M/EEG). However, it is still unclear to what extent speech-evoked event-related potentials (ERPs) including well-established phenomena, e.g., the N1 selective attention effect, contribute to the regression-based analyses. Here, we addressed this question by analyzing an EEG dataset obtained during a simple multispeaker selective attention task in which participants were cued to attend to only one of two competing speakers. Segmenting the ongoing EEG based on acoustic edges, we were able to replicate previous findings of event-related responses to speech in MEG data with particularly clear P1-N1-P2 complexes. Crucially, speech-evoked ERPs exhibited significant effects of attention in line with the auditory N1 effect. Comparing speech-evoked ERPs to the linear regression results revealed two major findings. First, temporal response functions (TRFs) obtained from forward modeling were strongly temporally as well as spatially correlated with corresponding true ERPs. Second, effects of attention demonstrated by the stimulus reconstruction (SR) accuracies obtained from backward modeling appeared to be driven by a consistent generation of speech-evoked ERPs including the N1 effect. Taken together, our observations reveal a direct link between ERPs to acoustic edges in speech and the linear TRF and SR modeling techniques. We emphasize the enhancement in signal-to-noise ratio provided by repeatedly evoked N1 responses to be a critical factor in facilitating the tracking and subsequent higher-order processing of selectively attended speech. In addition to that, the findings imply a cortical speech representation through superimposed speech-evoked ERPs in accordance with recent arguments promoting the neural evoked-response speech tracking model.