EEG study: the brain briefly tracks two speakers at once when switching attention

Researchers at Trinity College Dublin and collaborators recorded EEG from normal-hearing adults in a simulated cocktail-party setup, with two competing speech streams layered over background babble. Participants were cued to shift their attention between speakers every 15–30 seconds. Using Temporal Response Functions to decode which stream the brain was locking onto, the team found that attention switching is not a clean handoff: neural tracking of the new target speaker ramps up before the brain fully disengages from the old one, producing a brief window where cortex simultaneously encodes both streams. A drop in EEG alpha power tracked these transitions, offering a readout of the cognitive effort involved at each phase of the switch.

The more novel angle is on the language side. The authors isolated cortical signals tied to lexical prediction and tested four different strategies — built with large language models — for how listeners rebuild linguistic context after switching speakers. The data suggest listeners essentially reset their accumulated lexical context rather than carrying it over, implying the brain treats the new stream as a fresh conversation instead of splicing it onto the previous one.

Beyond the auditory-neuroscience result, the work is notable for using LLMs as a modeling tool for human language processing and for its fully open release: the preprocessed EEG dataset, analysis code, and speech stimuli are all published on Zenodo under a Creative Commons license. Practically, sharper models of how attention reallocates in multi-talker environments feed directly into next-generation hearing aids and attention-decoding interfaces, one of the paper’s funders being a hearing-technology foundation.