Simultaneous cortical tracking of competing speech streams during attention switching

Carta, Sara; Aličković, Emina; Zaar, Johannes; Valdés, Alejandro López; Di Liberto, Giovanni M. · 2025 · Crossref

DOI: 10.1101/2025.07.02.662762

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This study investigates the neural mechanisms underlying attention switching in multi-talker environments, addressing a gap in neurophysiology literature that has predominantly focused on sustained attention. The authors aimed to determine how the brain reallocates focus between competing speech streams, specifically examining whether engagement with a new stream and disengagement from the previous one occur symmetrically or asymmetrically. Additionally, the research explored how lexical context is updated during these switches, testing whether listeners reset their semantic priors or maintain context across streams. The experiment utilized electroencephalography (EEG) recordings from normal-hearing adults in an immersive setting. Participants listened to two competing TED talk streams presented via front-facing loudspeakers, accompanied by background babble from rear speakers. Visual cues instructed participants to switch their attention between the left and right streams every 15–30 seconds. Neural tracking of speech was quantified using Temporal Response Functions (TRFs), which modeled the linear relationship between speech features (envelope, word onset, and lexical surprisal/entropy) and EEG responses. To assess lexical context updating, the authors constructed four computational models using Large Language Models: a switch-unaware "Oracle" model, a "Speaker-Specific" model, an "Attention" model, and a "Reset" model that assumed context was cleared at each switch. The results demonstrated that attention switching involves asymmetric neural processes. Engagement with the newly attended speaker began and ended significantly earlier than disengagement from the previously attended speaker, resulting in a transient period where both streams were simultaneously encoded. This transition was closely linked to a reduction in EEG alpha power, with the minimum alpha power occurring significantly after the point of encoding switch. Regarding lexical processing, the "Reset" model, which assumed listeners discard previous context upon switching, yielded the highest EEG prediction correlations when using lexical entropy as a regressor. This model also produced lower TRF-N400 amplitudes compared to models that retained prior context, suggesting that listeners effectively reset their lexical predictions when shifting attention to a new speaker. These findings elucidate the temporal and contextual dynamics of auditory attention shifts, revealing that the brain prioritizes rapid engagement with new information while maintaining a brief overlap with the previous stream. The evidence for a lexical context reset implies that listeners treat each attention block as a distinct semantic unit, rather than integrating context across switched streams. This work provides critical insights into the brain’s capacity for flexible speech processing in complex listening environments, highlighting the distinct neural signatures of attentional disengagement and engagement.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.6-27b-nvidia on 2026-08-10; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• auditory