Trinity College study reveals brain encodes two speech streams during attention switches
New research published in PLoS Biology challenges assumptions about how the human brain manages competing auditory inputs, finding that neural tracking of a new speaker begins before disengagement from the previous one is complete.
A study published in PLoS Biology by researchers at Trinity College Dublin has revealed that the human brain can transiently encode two competing speech streams simultaneously when switching attention. Using electroencephalography (EEG) recordings from 24 normal-hearing adults in an immersive multi-talker environment, the research demonstrates that neural tracking of a new speaker emerges before disengagement from the previous target is fully complete. This asymmetric process suggests a flexible mechanism for auditory attention that allows for rapid reallocation of cognitive resources.
The experiment required participants to switch their attention between two foreground speech streams, presented as TED talks, every 15 to 30 seconds based on visual cues. Background babble noise from 16 talkers was played to simulate a complex listening scenario. The researchers utilised Temporal Response Functions to measure neural tracking, finding that engagement with a new stream starts and ends significantly earlier than disengagement from the previous one. This indicates a brief period where both streams are encoded in the cortex, supporting the ability to monitor alternative auditory sources while maintaining focus on a target.
This transition was closely mirrored by a reduction in EEG alpha power, specifically in the 8–12 Hz band, which correlates with listening effort and cognitive load. The minimum in alpha power occurred after the encoding switch point but before the disengagement process was fully complete. This finding extends prior literature linking alpha-band activity to listening effort, providing specific temporal dynamics for how cognitive demand shifts during attention reorientation in naturalistic settings.
The study also investigated how listeners update lexical context when switching attention. By comparing four context-accumulation models constructed using Large Language Models, the researchers found that a Reset model best predicted the EEG data. This model assumes that listeners ignore prior context and reset their lexical predictions upon switching attention. The results suggest that rather than maintaining semantic information from the previous stream, the brain recalibrates its linguistic expectations to align with the new target.
Behavioural data supported the neural findings, with participants achieving an average accuracy of 86.3% in answering content questions about the attended stream. Participants rated the difficulty of switching attention at 3.1 out of 5, indicating a moderate cognitive load. The study, supported by the William Demant Fonden and Research Ireland, offers new insights into the neural mechanisms of dynamic attentional reallocation, highlighting the brain’s capacity for flexible speech processing in complex environments.
