Figures
Abstract
Successful speech communication in multi-talker scenarios requires a skillful combination of sustained attention and rapid attention switching. While the neurophysiology literature offers detailed insights into the neural underpinnings of sustained attention, there remains considerable uncertainty on how attention switching takes place. In this study, using EEG recordings from normal-hearing adults in an immersive multi-talker environment, we measured the neural encoding of two competing speech streams amid background babble. Participants were cued to switch attention between streams every 15–30 s. Neural tracking was assessed via Temporal Response Functions (TRF), confirming reliable decoding of attentional focus. Our results indicate asymmetric disengagement and engagement processes during attention switches, where the neural tracking of the new target stream emerges before disengaging from the previous target, revealing a transient simultaneous encoding of two speech streams. That transition was closely mirrored by a reduction in EEG alpha power, informing on the cognitive effort during different phases of the attention switch. We then isolated cortical activity reflecting lexical prediction mechanisms to determine how lexical context is updated after an attention switch, comparing four context-accumulation strategies that were constructed using Large Language Models. Our findings elucidate both the temporal and contextual mechanisms underlying auditory attention shifts, pointing to the possibility that listeners carry out a reset in lexical context after switching attention. By focusing on dynamic attentional reallocation, this study offers insights into the brain’s capacity for flexible speech processing in complex listening environments.
Citation:
Carta S, Aličković E, Zaar J, López Valdés A, Di Liberto GM (2026) Competing speech streams are simultaneously represented in the human cortex during attention switching. PLoS Biol 24(7):
e3003876.
https://doi.org/10.1371/journal.pbio.3003876
Academic Editor:
Manuel S. Malmierca, Universidad de Salamanca, SPAIN
Received:
July 3, 2025;
Accepted:
June 12, 2026;
Published:
July 16, 2026
Copyright:
© 2026 Carta et al. This is an open access article distributed under the terms of the
Creative Commons Attribution License
, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Data Availability:
All data supporting the findings reported in this manuscript are freely accessible without restriction. The EEG pre-processed dataset, the resulting analysis files, and the analysis code are publicly available on the open repository Zenodo (
https://zenodo.org/records/20569817
). The EEG recordings are provided following the Continuous-event Neural Data (CND) format standard. The associated speech stimuli can also be found in the same repository, within the STIMULI folder.
Funding:
S.C., A.L.V., and G.D.L. were supported by the William Demant Fonden (
https://www.williamdemantfonden.dk/
), under grants 21-0628 and 22-0552, and by Taighde Éireann – Research Ireland (
https://www.researchireland.ie/
) under grant No. 18/CRT/6223. G.D.L. additionally conducted this research with the financial support of Research Ireland at ADAPT, the Research Ireland Centre for AI-Driven Digital Content Technology (
https://www.adaptcentre.ie/
) at Trinity College Dublin [grant 13/RC/2106_P2]. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing interests:
The authors have declared that no competing interests exist.
Abbreviations:
EEG,
electroencephalography; EOG,
electro-oculography; EMG,
electro-myography; ERSP,
event-related spectral perturbation; iEEG,
intra-cranial electroencephalography; FDR,
false discovery rate; fMRI,
functional magnetic resonance imaging; ICA,
Independent Component Analysis; IQR,
interquartile range; LLM,
large language model; MEG,
magnetoencephalography; PSD,
power spectral density; RMS,
root-mean-squared; SE,
standard error; SEM,
standard error of the mean; SNR,
signal-to-noise ratio; SPL,
sound pressure level; TRF,
Temporal Response Functions
Introduction
To understand speech in multi-talker environments, listeners single out the target speaker from competing sound streams [
1
–
3
]. The neurophysiology of this selective attention process has been widely studied with simulated
cocktail-party
scenarios [
4
,
5
], shedding light on how our brains segregate a target stream from competing speech streams, and enabling the transformation of the target speech into linguistic meaning. While the extent to which masker speech streams are processed remains highly debated [
6
–
8
], there is no doubt that there are considerable differences between the processing of target and masker speech, which have been measured with various technologies, such as non-invasive electroencephalography (EEG) [
1
,
9
], intra-cranial electroencephalography (iEEG) [
10
], magnetoencephalography (MEG) [
3
,
11
] and functional magnetic resonance imaging (fMRI) [
12
,
13
]. That work could pinpoint precise loci in the auditory cortical areas where that segregation emerges [
14
] as well as measuring the substantial (but not total) suppression of linguistic processing for the masker speech [
1
,
15
–
17
]. However, neurophysiology literature in this field has almost entirely focused on sustained attention tasks [
2
,
10
], leaving considerable uncertainty on the neural underpinnings of attention switching.
Dynamic switching paradigms have been widely used in the domain of cognitive control studies to probe for cognitive flexibility and cognitive stability [
18
]. In those experiments, participants are often required to flexibly adapt their behavioral response depending on new instructions, initiating a task-switch [
19
–
21
]. For example, given a single digit, they are required to classify it either based on parity, i.e., whether it is even or odd, or based on relative magnitude, i.e., whether the digit is greater than or less than 5 [
22
]. In these paradigms, the switch-cost is the increase in reaction time or error rate when switching from one task to the other. Similar behavioral paradigms have also involved simple speech stimuli in multi-talker settings [
23
–
25
]. However, the main interest of those tightly controlled experiments was to model the process of target speech selection as one particular instance of a task-switching problem, i.e., target stream selection could either depend on spatial location or voice identity [
23
], rather than focusing on the dynamic aspect of attention re-allocation per se in naturalistic multi-talker scenarios. As such, very little is known on how a flexible reorienting of attention might impact speech processing of continuous competing streams.
In recent speech neurophysiology research, experimental paradigms have started to include switches of attention as a tool towards tailored EEG/MEG methodological advances in the domain of attention decoding [
26
,
27
], or to investigate how sustained speech attention unfolds for moving auditory objects [
28
]. However, to the best of our knowledge, only one previous study has specifically focused on the neurophysiology of attention switching in multi-talker scenarios, relating the neural encoding of speech during attentional re-orienting with EEG alpha activity and pupil dilation dynamics [
29
]. Those findings proved that the neurophysiology of attention switching can be studied non-invasively. Building on that work, our study sheds light on the exact neural dynamics supporting the steering of attention between two competing speech streams, disengaging from the previous target stream while engaging to the new one.
In this study, we measure the neural encoding of speech using a range of encoding window lengths, as listeners steer their attention from one speaker to another. We test whether engagement with a new speech stream begins before disengagement from the previous target is complete, resulting in a brief period of simultaneous tracking of both streams. Such an asymmetry in the disengagement-engagement processes, even if transient, could support the ability to explore alternative auditory streams while maintaining attention to a given stream [
30
].
The neural encoding of speech was measured from normal-hearing adult participants using EEG during an immersive multi-talker listening task. Participants were exposed to two competing speech streams from TED talks, presented via two front-facing loudspeakers, while background noise from a 16-talker speech babble played from rear loudspeakers (
Fig 1A
). An on-screen arrow cued participants to attend to one of the two speech streams and to shift their attention rapidly whenever the arrow changed direction, approximately every 10–30 s (
Fig 1B
). Neural tracking of target and masker speech was quantified using the Temporal Response Function (TRF), describing the linear relationship between each speech stream and the neural responses. As an initial validation, we confirmed that the attended stream could be reliably decoded from the EEG, consistent with the extensive literature on sustained attention [
9
,
10
,
31
]. This confirms that the EEG responses in this experiment reflects differential encoding of target versus masker speech (
Fig 1C
).
Download:
PNG
larger image
TIFF
original image
Fig 1.
Experiment overview and validation.
(A)
Participants were presented with speech from two loudspeakers placed in front of them with 60° of separation (30° left and 30° right), and with concurrent 16-talker background noise (B1–B4). In each trial, the screen presented an arrow pointing to the target speech stream. Participants were instructed to switch attention as soon as the visual cue changes direction.
(B)
Schematic diagram of one experimental trial. The black area represents blocks of attention either to the left (L) or right (R) front streams. The red arrows indicate the instants where the attention cue switches side (six times per trial). Note that block duration was randomized and always between 15 and 30 s, with trials lasting 3 min.
(C)
EEG data validation was carried out by running an attention decoding analysis. Progressively longer decoding windows were considered (larger windows use more data, typically leading to more accurate decoding scores). Binary classification scores are reported arbitrating between the target and masker streams. The dashed line indicates the 95th percentile of a random distribution calculated by randomizing the classification labels. Statistically significant attention decoding classification scores were measured for all the decoding windows considered, with numerical results comparable with previous studies on selective attention [
31
,
34
,
35
]. Data supporting this figure is available at:
https://zenodo.org/records/20569817
.
https://doi.org/10.1371/journal.pbio.3003876.g001
We next addressed two fundamental questions about the neural mechanisms underlying attention switching in naturalistic listening. First, we asked whether the processes of engaging with a new speech stream and disengaging from a previous one unfold symmetrically (
Figs 2
and
3
). To test this, we fit encoding TRF models to EEG data, measuring the neural tracking of the two competing speech streams over time. This allowed us to characterize the average encoding dynamics surrounding attention switches, comparing disengagement and engagement processes. The second objective was to understand how our brains update and use lexical context when switching attention (
Fig 4
). Building on previous work showing that speech comprehension is supported by contextual predictions [
32
,
33
], we formulated four competing hypotheses reflecting different assumptions about how linguistic context is preserved, reset, or selectively updated across an attention switch. Using a state-of-the-art large language model (LLM), we derived quantitative predictions for each hypothesis, resulting in four regressors for lexical surprisal and entropy, separately, differing in their sensitivity to prior context and to the occurrence of the switch. Encoding TRF models were then fit for each hypothesis, allowing us to compare alternative context-accumulation strategies and identify the model most consistent with the observed neural responses. This study provides substantial new insights into the temporal unfolding and contextual mechanisms guiding attention switching, encompassing both low and high levels of speech abstraction.
Download:
PNG
larger image
TIFF
original image
Fig 2.
The attention-switching cue prompts a robust disengagement from Speaker 1 and engagement to Speaker 2, and it is followed by a significant decrease in the EEG alpha ERSP.
Disengagement has longer temporal dynamics compared to engagement.
(A) Left:
Speech tracking encoding for an attention switch from Speaker 1 and 2. The trajectory in the panel represents our null hypothesis, where the disengagement and engagement processes progress in a symmetric manner after the switch-cue (vertical gray line).
Right:
Results for the neural tracking of Speaker 1 and Speaker 2 across the switching cue. EEG prediction correlations (average across all channels) obtained from a 4-s sliding-window TRF model including Envelope (Env), Word Onset (WO) and Word Surprisal (WS) features. Coloured horizontal bars at the bottom of the plot indicate the attention instruction around the attention switching cue. The turquoise dot indicates the encoding switch of EEG prediction correlations based on Spk1- and Spk2- speech features. The piecewise linear model fit for disengagement and engagement is overlayed on the EEG prediction correlation values. Please note that the broken-line-fit in this plot was performed on the grand-average cortical tracking curves here for illustrative purposes. Please find the estimates at the single-participant level in Panel
C.
Hexagram shapes indicate the start of the disengagement (blue) and engagement (yellow) processes, while diamonds represent the end of the transitions.
(B) Left:
Diagram of expected results for alpha-band ERSP (event-related spectral perturbation) across the switching cue.
Right:
ERSP of the alpha band (8–12 Hz) around the switching cue (average of all channels), computed with a 4-s sliding window, as above. Scalp topographies at selected time points reveal a pattern of posterior negativity, which drops significantly following the instruction to switch (thick black lines indicate a statistically significant change compared to pre-switch baseline). The red dot represents the average of ERSP minima across participants. The shaded area represents the standard error of the mean (SEM) across participants.
(C) Left:
Comparison of encoding switch of EEG prediction correlations (turquoise bar) and alpha ERSP minimum (red bar) for a 4-s sliding window. The alpha ERSP reaches its minimum significantly after the Spk1-Spk2 encoding switch point.
Right:
Comparison of temporal dynamics for start and end points of disengagement and engagement processes, with start/end transition points estimated at the single-participant level. Stars indicate significant statistical effects (paired sample
t
-tests; *
p
≤ 0.05; **
p
≤ 0.01; **
p
≤ 0.001). Data supporting this figure is available at:
https://zenodo.org/records/20569817
.
https://doi.org/10.1371/journal.pbio.3003876.g002
Download:
PNG
larger image
TIFF
original image
Fig 3.
Comparing the start and end transition points for the disengagement and engagement processes after the attentions switching cue.
The process of engaging to a new speaker begins and ends significantly earlier than disengaging from the previously attended speaker.
(A, B)
Start and end points of the transition for the disengagement (blue) and engagement (yellow) processes over five TRF sliding window lengths. Error bars represent SEM across participants. Stars indicate significant effects of process type (two-way repeated measures ANOVA; *
p
≤ 0.05; **
p
≤ 0.01; ***
p
≤ 0.001). Data supporting this figure is available at:
https://zenodo.org/records/20569817
.
https://doi.org/10.1371/journal.pbio.3003876.g003
Download:
PNG
larger image
TIFF
original image
Fig 4.
Investigating lexical prediction mechanisms during attention switching.
(A)
Layout of the four context models. Blocks coloured in black illustrate sustained attention either to the Left or Right stream, while orange arrows indicate attention switching cues. The thick red arrow indicates the context used to guide word predictions for the current block (B7, highlighted in orange).
(B)
Average lexical entropy at words preceding and following the attention switch cue. Note that no value for entropy is displayed in the Reset model for the first word after the switch, due to the context being fully reinstated.
(C)
EEG prediction correlations for the four multivariate TRF models, only differing in their entropy feature. Coloured dots indicate the average across all electrodes and participants. The gray area at the bottom represents the average encoding accuracy of a multivariate TRF without any semantic information (Envelope + Word Onset). Stars represent statistically significantly greater EEG prediction correlations for the Reset model compared to the other models (Significance levels: *
p
< 0.05, **
p
< 0.01, ***
p
< 0.001). Topographical patterns illustrate the gain due to semantic information (compared to the Envelope + Word Onset TRF) for the four models.
(D)
(Left) TRF weights for the entropy feature at time-lags between −100 and 600 ms relative to stimulus onset. Transparent shaded areas represent the standard error of the mean (SEM) across participants. The horizontal black line indicates the time window employed to compute the average TRF-N400 amplitude. (Right) Boxplots representing the distribution of the TRF-N400 amplitude across participants for the four context models. The central line within each box represents the median, while the edges of the box indicate the interquartile range (IQR). Whiskers extend to the most extreme data points within 1.5 times the IQR from the quartiles. Outliers are plotted as individual points beyond the whiskers. Stars indicate statistically significant differences (Significance levels: *
p
< 0.05, **
p
< 0.01, ***
p
50%) over the course of the switch. In doing so, we removed participants for whom the start and end points of the (dis)engagement could not be estimated (note that this exclusion is determined before identifying the start/end estimates; in that sense, this is different from an outlier removal, which would exclude extreme start/end values instead).
Aligning with our expectation (
Fig 2A
), EEG prediction correlations around the switching cue reflected tracking of Spk1 and Spk2 streams consistent with the attention instructions, such that Spk1 was significantly more tracked than Spk2 before the switch, while the reverse pattern was observed after the switch (paired two-tailed
t
test of Spk1-Spk2 difference against zero, FDR-corrected for multiple comparisons,
p
< 0.005).
As the attention switch unfolds, also the grand-mean ERSP in the alpha frequency band displayed a statistically significant change compared to baseline (one-sample
t
test against zero, FDR-corrected for multiple comparisons), revealing a pattern of occipito-parietal negativity in the scalp topographies (
Fig 2B
). This is consistent with our expectation of an impact of attentional reorientation on the EEG alpha band, which has already been shown to reflect attention switching behavior in competing speech listening scenarios [
29
].
EEG prediction correlations for Spk1 and Spk2 converged, before significantly separating again once the switching process was concluded and, presumably, the attention was fully reallocated. Here, we refer to the time point when EEG prediction overlaps between Spk1 and Spk2 as the encoding switch point. Given the observed statistically significant drop in alpha ERSP, we asked how the temporal dynamics of this drop compared to those of the EEG prediction correlations. To address this, for each participant, we identified the time of the alpha ERSP minimum, and the encoding switch point, based on an encoding window of 4s (
Fig 2C
). The choice of this particular encoding window for our main analysis is justified based on the classification accuracy results (
Fig 1C
), since it is a good compromise between temporal resolution and classification performance. However, the same pattern of results holds when considering multiple encoding windows simultaneously (
S1 Fig
). A paired
t
test comparing the temporal dynamics of the alpha ERSP and the EEG prediction correlations showed that the minimum of the alpha ERSP drop significantly follows the encoding switch point (
t
(20) = 4.29,
p
= 3.59e-4, Cohen’s
d
= 0.94).
We then evaluated multiple encoding window lengths, assessing the effect of Metric (encoding switch versus ERSP minimum) and Window (1, 2, 4, 8 s) on the timing of the encoding switch and the minimum of the alpha ERSP with a 2-way repeated measures ANOVA. The analysis revealed that the temporal dynamics of both encoding switch and alpha ERSP minimum became longer as the encoding window length increased (
F
(1.68, 33.55) = 52.77,
p
= 2.3e−10,
η
p
2
= 0.72; a Greenhouse-Geisser’s correction applied due to sphericity violation), which is unsurprising given the methodological constraints we discussed (see
Methods
). More interestingly for our question, a statistically significant effect of Metric emerged (
F
(1,20) = 20.26,
p
= 2.18e−4,
η
p
2
= 0.5), with the alpha ERSP minimum occurring significantly later than the encoding switch across a range of encoding windows (Holm-corrected post-hoc
t
test:
t
(20) = 4.5,
p
= 2.18e−4, Cohen’s
d
= 0.97).
Dissecting the temporal dynamics of attentional disengagement and engagement during attention switching
The attention switching cue prompts the listener to reallocate their attention from the previously attended speaker, Spk1, to the newly attended speaker, Spk2. While this re-routing of attention appears to be a single, unified process, it is possible to distinguish two separate operations that are necessary for it to happen: disengagement, which we define as the decrease in neural tracking for the previously attended speech stream, and engagement, which we define as the increase in neural tracking for the previously unattended speech stream. Our goal was to clarify the temporal dynamics of these two operations to understand whether they occur fully in parallel, serially, or with a certain degree of overlap. It is worth noting that, due to the use of sliding encoding windows, the estimated temporal dynamics of engagement and disengagement do not reflect the exact time course of the underlying neural processes and should be interpreted as relative, rather than absolute, temporal metrics. As in the previous analysis, we first selected participants displaying a reliable attentional bias over the course of the switch (see
Methods
). For this selection of 21 participants, we fitted a piecewise linear regression on single-subject EEG prediction correlations, and found the optimal breakpoints, corresponding to the start and end time points of disengagement and engagement (
Fig 2C
). As above, we chose to focus on an example window of 4 s and later replicated our results on a range of encoding window lengths. To further characterize the spatial patterns of engagement and disengagement processes, scalp topographies of the EEG prediction correlations at selected time points are shown in
S2 Fig
, indicating that the most predictive channels were predominantly located over central-parietal regions. Disengagement and engagement processes were compared separately based on their start times and end times, revealing consistently earlier temporal dynamics for the engagement compared to the disengagement. Engagement to the newly attended speaker started significantly earlier than the disengagement from the previously attended speaker (paired-sample
t
test:
t
(20) = 2.37,
p
= 0.03, Cohen’s
d
= 0.52), and finished significantly earlier (paired-sample
t
test:
t
(20) = 2.35,
p
= 0.03, Cohen’s
d
= 0.39).
We then extended our analysis to a range of sliding window lengths and compared start times and end times for disengagement and engagement processes, including Window (1, 2, 4, 8, 16 s) and Process (disengagement versus engagement) as main factors in a repeated measures ANOVA. Regarding the start points (
Fig 3A
), our analyses revealed an expected statistically significant effect of Window (
F
(1.51,30.14) = 9.7,
p
= 0.001,
η
p
2
= 0.33; the assumption of sphericity was not met; hence, a Greenhouse–Geisser’s correction was applied), with longer temporal dynamics corresponding to longer encoding window lengths. More importantly, we also observed a significant main effect of Process (
F
(1,20) = 5.48,
p
= 0.03,
η
p
2
= 0.21), with engagement to the newly attended stream starting significantly earlier than the disengagement to the previously attended stream (Holm-corrected post-hoc
t
test:
t
(20) = 2.34,
p
= 0.03, Cohen’s
d
= 0.54). The same statistical analysis was repeated separately on the end time points of disengagement and engagement processes (
Fig 3B
), revealing once again a main effect of Window, whereby longer encoding windows yield longer temporal transitions (
F
(1.97,39.32) = 31.76,
p
= 7.2e−9,
η
p
2
= 0.61; the assumption of sphericity was not met; hence, a Greenhouse–Geisser’s correction was applied). A significant main effect of Process also emerged (
F
(1,20) = 4.46,
p
= 0.047,
η
p
2
= 0.18), revealing that the process of engagement to the newly attended speaker, not only starts, but also ends significantly earlier than the disengagement (Holm-corrected post-hoc
t
test:
t
(20) = 2.11,
p
= 0.047, Cohen’s
d
= 0.58).
A follow-up analysis including the three participants with lower-than-chance classification accuracy around the switching cue confirmed that these data points introduced noise to the estimation of engagement and disengagement latencies. This was expected, as the start and end transition points cannot be determined in those participants. The patterns observed were qualitatively similar to the main result reported above, with earlier temporal dynamics for the engagement compared to the disengagement, albeit with weaker effects below the statistical significance threshold (repeated-measures ANOVA; start point:
F
(1,23) = 2.69,
p
= 0.11,
η
p
2
= 0.1; end point:
F
(1,23) = 2.96,
p
= 0.1,
η
p
2
= 0.11).
Determining how lexical predictions are built during attention switching
Reorienting attention to a different speech stream implies a change of context and, consequently, different semantic priors for lexical predictions. We thus hypothesized that incorporating this change of context into the structure of our semantic regressor in a multivariate encoding TRF model would increase EEG prediction correlations, as it would better reflect the dynamically updating neural tracking of the competing speech streams. We compared four alternative models representing how context could be incrementally accumulated for performing lexical predictions at one particular attention block (e.g., B7, in
Fig 4A
). A naïve Oracle model, which uses all available context of previous blocks from the current stream, whether attended or unattended, to predict words from the current block, served as our baseline, since it was essentially a switch-unaware contextual representation. Speaker-Specific and Attention models were instead switch-aware models, as they only considered previously attended blocks as part of the context for lexical predictions. Speaker-Specific assumed a higher degree of stream segregation, since its context only consisted of previously attended blocks from the same speech stream, while Attention included any previously attended block from both streams. The Reset model instead ignored all previously attended blocks from any of the streams and computed context only over the course of the current block of attention, as if the priors for lexical predictions were reset at each attention switch (
Fig 4A
).
As lexical entropy is a proxy of uncertainty for next-word prediction, its values should be impacted by a switching cue, which determines an abrupt change of context.
Fig 4B
shows the change of average lexical entropy values in words preceding and following the switch cue, which vary depending on the context models. It can be observed that the Reset model peaks with the highest uncertainty and slowly decays over the course of the next words, while the Attention and Speaker Specific models have overall similar lexical entropy dynamics and more stable values. Consistently with its switch-unaware nature, the Oracle model instead displays entropy values that are largely unchanged despite the switch. An explicit comparison of the average entropy values of the four context-accumulation strategies revealed statistically significant differences (repeated-measures ANOVA, Greenhouse-Geisser correction due to sphericity violation
; F
(1,19) = 39.57,
p
= 9.59e−10,
η
p
2
= 0.68). Post-hoc pairwise tests (Holm-adjusted) indicated that the Reset model showed an intermediate average entropy, significantly higher than the Oracle model (
t
(19) = 5.75,
p
= 1.44e−6, Cohen’s
d
= 0.45), and significantly lower than the Attention (
t
(19) = 4.64,
p
= 6.28e − 5, Cohen’s
d
= 0.36) and Speaker-Specific (
t
(19) = 2.24,
p
= 0.04, Cohen’s
d
= 0.17) models. As such, despite showing the highest peak following the attention switching cue, the Reset model had overall intermediate entropy values across the four lexical expectation models considered here.
Lexical surprisal and lexical entropy were used as semantic information regressors for each context model and separately included in a multivariate stimulus representation to fit single-subject encoding TRFs (Envelope-Word Onset-Word Surprisal and Envelope-Word Onset-Word Entropy). Resulting TRF weights and EEG prediction correlations were then compared across context models, with the hypothesis that switch-aware and context-rich representations (e.g., Speaker-Specific or Attention) would best describe neural activity in attention-switching scenarios.
Before comparing the context models, we first tested whether each of them yielded a significant encoding accuracy gain compared to the baseline model only consisting of acoustic features (Envelope and Word Onset). When using entropy as a regressor for semantics, all models, with the exception of Oracle, showed a statistically significant gain, suggesting a robust tracking of semantic information in addition to the stimulus acoustics (paired t-tests: Oracle versus Acoustics:
p
= 0.2; Spk.Spec. versus Acoustics:
p
= 0.04; Attention versus Acoustics:
p
= 0.04; Reset versus Acoustics:
p
= 0.002). Employing word surprisal as semantic regressor yielded similar results, with all the models showing a robust encoding of semantic information, apart from Oracle (paired t-tests: Oracle versus Acoustics:
p
= 0.15; Spk.Spec. versus Acoustics:
p
= 0.02; Attention versus Acoustics:
p
= 0.02; Reset versus Acoustics:
p
= 0.01). The non-significant gain of the Oracle model compared to the acoustic model was expected, since Oracle was designed as a control
switch-unaware
model.
In contrast to our expectation, the Reset context model was shown to yield higher EEG prediction correlation values when entropy was used as a regressor for semantics (
Fig 4C
). A repeated measures ANOVA revealed a statistically significant effect of the main factor, Context Model (
F
(2.1,47.75) = 9,
p
= 4e−4,
η
p
2
= 0.28; with Greenhouse-Geisser’s correction). In the Holm-corrected post-hoc tests, the Reset model was shown to yield significantly higher encoding accuracies than Oracle (
t
(23) = 4.99,
p
= 2.63e−5, Cohen’s
d
= 0.14), Speaker Specific (
t
(23) = 3.73,
p
= 0.002, Cohen’s
d
= 0.1), and Attention (
t
(23) = 3.28,
p
= 0.006, Cohen’s
d
= 0.09). We then assessed the difference of TRF weights for the entropy feature across the four context models (
Fig 4D
), averaging the weights’ amplitude within a window broadly centered around the TRF-N400 latency (350–550 ms). A repeated measure ANOVA was run on the weights’ amplitude values, revealing a main effect of Context Model (
F
(3,69) = 15.51,
p
= 8.2e−8,
η
p
2
= 0.4). Post-hoc tests (Holm-corrected) showed that weights for the Reset model had lower TRF-N400 amplitude compared to Oracle (
t
(23) = −5.56,
p
= 2.4e−6, Cohen’s
d
= 0.45), Attention (
t
(23) = −5.84,
p
= 9.2e−7, Cohen’s
d
= 0.47), and Speaker Specific (
t
(23) = −5.24,
p
= 6.5e−6, Cohen’s
d
= 0.43).
When fitting a multivariate TRF including lexical surprisal as a semantic regressor, we observed a statistically significant difference in EEG prediction correlations between the four context models (
F
(1.59,36.6) = 3.96,
p
= 0.04,
η
p
2
= 0.15, with Greenhouse–Geisser correction). Post-hoc analyses indicated a statistically significant difference between the Reset and Oracle models (
t
(23) = 3.18,
p
= 0.013, Cohen’s
d
= 0.1), while all other post-hoc pairwise comparisons did not reach the significance threshold (
p
unattended) and engaging (unattended > attended) stream. To compare the onset and offset points of these processes, we first selected participants depending on their attentional bias over the course of the switch. Based on the reference 4-s encoding window, we determined the single-subject classification accuracy in a 15 s window before and after the attention switching cue. We then found those participants displaying a lower-than-chance (50%) classification performance either in the pre- or post-switch window and excluded them from further analyses on the switching temporal dynamics. Start and end points of disengagement and engagement processes were identified for each subject by fitting a piecewise linear regression model to the EEG prediction correlation values for the disengaging and the engaging streams, around the attention switch cue. For each of the EEG prediction correlation streams, the model incorporates the average pre-switch and post-switch values as constants and tries to estimate the breakpoints at which these flat segments connect, i.e., the onset and offset points of the process. An initial guess for the breakpoints was provided based on reasonable assumptions (e.g., that the start of the disengagement/engagement could not occur before the switching cue), and the fitting was done with MATLAB’s
lsqcurvefit
function, which minimizes the difference between the model and the observed data.
When computing encoding TRFs comparing different kinds of context-accumulation models, we adopted an approach optimized for highlighting the effects of lexical information processing [
83
]. With this method, the prediction correlation between true and predicted EEG data is computed within a window covering the range from 300 to 500 ms after word onset, allowing to effectively enhance the impact of semantic encoding.
Event-related spectral perturbation (ERSP)
Previous studies have related decreased parietal alpha-band (8–12 Hz) EEG power to increased listening effort and have shown that more listening effort is required to switch attention than to sustain attention on a particular speaker [
29
]. Since we were interested in assessing the time course of alpha power and how it would compare to the temporal dynamics of the TRF prediction correlations, we computed event-related spectral perturbation (ERSP) in the alpha frequency band as a proxy of effort around the switch. Since neural activity in the alpha frequency band is sensitive to ocular artifacts, we performed Independent Component Analysis (ICA) using standard EEGLAB functions [
84
]. For each subject, we identified independent components reflecting ocular artifacts (such as blinks and horizontal eye movements) based on their temporal characteristics and spatial topographies, and we removed them from the data. ICA-based artifact removal was applied only for the alpha-band ERSP analysis and not for neural tracking. TRF estimates are based on stimulus–response covariance over extended time lags and are therefore unlikely to be influenced by brief eye movements around the switch, which do not systematically covary with the speech signal. TRFs were fit on the full dataset, and inspection of the TRF weights did not reveal near-zero latency spikes indexing ocular contamination. Although eye movements could in principle affect the SNR around the switch, we did not observe corresponding changes in EEG prediction correlations (e.g.,
Fig 4B
, Oracle model), and switch direction was counterbalanced across left and right streams.
For each participant, trial, and attention switch, ERSP was computed as the relative change in power spectral density (PSD) compared to a baseline. We first defined a segment (±15 s) around the switching cue and, to estimate ERSP with the same temporal resolution used for the neural tracking analysis, we chose a sliding window of 4 s, with a step size of 0.125 s. This analysis was also replicated across a range of sliding window lengths (1, 2, 8 s), in line with the neural tracking analysis. Along the switching segment, PSD was calculated at each forward step within the sliding window via Welch’s method (MATLAB’s
pwelch
function) and summed across the alpha band frequency bins (8–12 Hz). The PSD computed at each step was compared to the reference PSD value of a baseline, consisting of the first sliding window in the pre-switch segment, producing a series of ERSP values (in dB) that reflects the relative change in alpha power during the switch.
Statistical analysis
Statistical analyses were performed using repeated measures ANOVAs with two or three factors, with
F
-values reported as
F
(df
time
, df
error
). When the assumption of sphericity wasn’t met, as revealed by Mauchly’s test of sphericity, a correction was applied with the Greenhouse–Geisser’s method. When a significant main effect was found, post-hoc tests were run, and
p
-values were corrected for multiple comparisons with the Holm method. To compare the significance of a time series against zero, a Wilcoxon signed rank test was used for each time point, and the false discovery rate (FDR) was controlled with the Benjamini and Hochberg procedure. We report FDR-adjusted
p
-values. Descriptive statistics for the neurophysiology results are reported as a combination of mean and standard error (SE).
Supporting information
S1 Fig.
Time latencies of the encoding switch of EEG prediction correlations (turquoise) and alpha ERSP minimum (red) for four sliding window lengths.
Bars represent the average time latencies across participants and electrodes; error bars indicate the SEM across participants. Stars represent statistically significant comparisons (Significance levels: *
p
< 0.05, **
p
< 0.01, ***
p
< 0.001). Data supporting this figure is available at:
https://zenodo.org/records/20569817
.
https://doi.org/10.1371/journal.pbio.3003876.s001
(TIF)
S2 Fig.
Neural tracking time courses and scalp topographies for Speaker 1 and Speaker 2 across the attention-switch cue.
EEG prediction correlations result from a model including Envelope, Word Onset, and Word Surprisal features using a sliding window of 4s. Scalp topographies indicate the predictive performance of the TRF model across electrodes at selected time points (black dots), for the engagement and disengagement processes. Coloured horizontal bars indicate the instructed focus of attention before and after the switching cue. The turquoise dot marks the encoding switch, when EEG tracking shifts from Spk1 to Spk2. Overlaid piecewise linear fits represent disengagement from the initially attended speaker and engagement with the newly attended speaker. Blue and yellow hexagrams denote the onset of disengagement and engagement, respectively, while diamonds indicate the end of each transition. Fits are shown on grand-average tracking curves for illustration; participant-level estimates are reported in
Fig 2C
. Data supporting this figure is available at:
https://zenodo.org/records/20569817
.
https://doi.org/10.1371/journal.pbio.3003876.s002
(TIF)
Acknowledgments
We thank ADAPT, the Research Ireland Centre for AI-Driven Digital Content Technology, and the Eriksholm Research Centre, part of Oticon, for the support. We thank Amirhossein Chalehchaleh, Emily Ip, and Martin Winchester for their help with data collection, and Dr. Enda Bates for assistance in facilitating access to the experimental space. For the purpose of Open Access, the author has applied a CC BY public copyright licence to any Author Accepted Manuscript version arising from this submission.
References
1.
Power AJ, Foxe JJ, Forde E-J, Reilly RB, Lalor EC. At what time is the cocktail party? A late locus of selective attention to natural speech. Eur J Neurosci. 2012;35(9):1497–503. pmid:22462504
View Article
PubMed/NCBI
Google Scholar
2.
Zion Golumbic EM, Ding N, Bickel S, Lakatos P, Schevon CA, McKhann GM, et al. Mechanisms underlying selective neuronal tracking of attended speech at a “cocktail party”. Neuron. 2013;77(5):980–91. pmid:23473326
View Article
PubMed/NCBI
Google Scholar
3.
Puvvada KC, Simon JZ. Cortical representations of speech in a multitalker auditory scene. J Neurosci. 2017;37(38):9189–96. pmid:28821680
View Article
PubMed/NCBI
Google Scholar
4.
Cherry EC. Some experiments on the recognition of speech, with one and with two ears. J Acoust Soc Am. 1953;25(5):975–9.
View Article
Google Scholar
5.
Bronkhorst AW. The cocktail-party problem revisited: early processing and selection of multi-talker speech. Atten Percept Psychophys. 2015;77(5):1465–87. pmid:25828463
View Article
PubMed/NCBI
Google Scholar
6.
Rivenez M, Darwin CJ, Guillaume A. Processing unattended speech. J Acoust Soc Am. 2006;119(6):4027–40. pmid:16838545
View Article
PubMed/NCBI
Google Scholar
7.
Har-Shai Yahav P, Zion Golumbic E. Linguistic processing of task-irrelevant speech at a cocktail party. Elife. 2021;10:e65096. pmid:33942722
View Article
PubMed/NCBI
Google Scholar
8.
Kaufman M, Zion Golumbic E. Listening to two speakers: capacity and tradeoffs in neural speech tracking during Selective and Distributed Attention. Neuroimage. 2023;270:119984. pmid:36854352
View Article
PubMed/NCBI
Google Scholar
9.
O’Sullivan JA, Power AJ, Mesgarani N, Rajaram S, Foxe JJ, Shinn-Cunningham BG, et al. Attentional selection in a cocktail party environment can be decoded from single-trial EEG. Cereb Cortex. 2015;25(7):1697–706. pmid:24429136
View Article
PubMed/NCBI
Google Scholar
10.
Mesgarani N, Chang EF. Selective cortical representation of attended speaker in multi-talker speech perception. Nature. 2012;485:233–6.
View Article
Google Scholar
11.
Simon JZ. The encoding of auditory objects in auditory cortex: insights from magnetoencephalography. Int J Psychophysiol. 2015;95(2):184–90. pmid:24841996
View Article
PubMed/NCBI
Google Scholar
12.
Evans S, McGettigan C, Agnew ZK, Rosen S, Scott SK. Getting the cocktail party started: masking effects in speech perception. J Cogn Neurosci. 2016;28(3):483–500. pmid:26696297
View Article
PubMed/NCBI
Google Scholar
13.
Hausfeld L, Hamers IMH, Formisano E. FMRI speech tracking in primary and non-primary auditory cortex while listening to noisy scenes. Commun Biol. 2024;7(1):1217. pmid:39349723
View Article
PubMed/NCBI
Google Scholar
14.
O’Sullivan J, Herrero J, Smith E, Schevon C, McKhann GM, Sheth SA, et al. Hierarchical encoding of attended auditory objects in multi-talker speech perception. Neuron. 2019;104(6):1195-1209.e3. pmid:31648900
View Article
PubMed/NCBI
Google Scholar
15.
Teoh ES, Ahmed F, Lalor EC. Attention differentially affects acoustic and phonetic feature encoding in a multispeaker environment. J Neurosci. 2022;42(4):682–91. pmid:34893546
View Article
PubMed/NCBI
Google Scholar
16.
Park H, Gross J. Get the gist of the story: neural map of topic keywords in multi-speaker environment. eLife Sciences Publications, Ltd. 2023.
View Article
Google Scholar
17.
Ding N, Simon JZ. Neural coding of continuous speech in auditory cortex during monaural and dichotic listening. J Neurophysiol. 2012;107(1):78–89. pmid:21975452
View Article
PubMed/NCBI
Google Scholar
18.
Egner T. Principles of cognitive control over task focus and task switching. Nat Rev Psychol. 2023;2(11):702–14. pmid:39301103
View Article
PubMed/NCBI
Google Scholar
19.
Luo T, Xu M, Zheng Z, Okazawa G. Limitation of switching sensory information flow in flexible perceptual decision making. Nat Commun. 2025;16(1):172. pmid:39747100
View Article
PubMed/NCBI
Google Scholar
20.
Ravizza SM, Carter CS. Shifting set about task switching: behavioral and neural evidence for distinct forms of cognitive flexibility. Neuropsychologia. 2008;46(12):2924–35. pmid:18601941
View Article
PubMed/NCBI
Google Scholar
21.
Wager TD, Jonides J, Reading S. Neuroimaging studies of shifting attention: a meta-analysis. Neuroimage. 2004;22(4):1679–93. pmid:15275924
View Article
PubMed/NCBI
Google Scholar
22.
Rogers RD, Monsell S. Costs of a predictible switch between simple cognitive tasks. J Exp Psychol General. 1995;124(2):207–31.
View Article
Google Scholar
23.
Lawo V, Fels J, Oberem J, Koch I. Intentional attention switching in dichotic listening: exploring the efficiency of nonspatial and spatial selection. Q J Exp Psychol (Hove). 2014;67(10):2010–24. pmid:25248101
View Article
PubMed/NCBI
Google Scholar
24.
Oberem J, Koch I, Fels J. Intentional switching in auditory selective attention: Exploring age-related effects in a spatial setup requiring speech perception. Acta Psychol (Amst). 2017;177:36–43. pmid:28456098
View Article
PubMed/NCBI
Google Scholar
25.
Meister H, Wenzel F, Gehlen AK, Kessler J, Walger M. Static and dynamic cocktail party listening in younger and older adults. Hear Res. 2020;395:108020. pmid:32698114
View Article
PubMed/NCBI
Google Scholar
26.
Miran S, Akram S, Sheikhattar A, Simon JZ, Zhang T, Babadi B. Real-time tracking of selective auditory attention from M/EEG: a bayesian filtering approach. Front Neurosci. 2018;12:262. pmid:29765298
View Article
PubMed/NCBI
Google Scholar
27.
Akram S, Presacco A, Simon JZ, Shamma SA, Babadi B. Robust decoding of selective auditory attention from MEG in a competing-speaker environment via state-space modeling. Neuroimage. 2016;124(Pt A):906–17. pmid:26436490
View Article
PubMed/NCBI
Google Scholar
28.
Teoh ES, Lalor EC. EEG decoding of the target speaker in a cocktail party scenario: considerations regarding dynamic switching of talker location. J Neural Eng. 2019;16(3):036017. pmid:30836345
View Article
PubMed/NCBI
Google Scholar
29.
Haro S, Rao HM, Quatieri TF, Smalt CJ. EEG alpha and pupil diameter reflect endogenous auditory attention switching and listening effort. Eur J Neurosci. 2022;55(5):1262–77. pmid:35098604
View Article
PubMed/NCBI
Google Scholar
30.
Agmon G, Yahav PH-S, Ben-Shachar M, Golumbic EZ. Attention to speech: mapping distributed and selective attention systems. Cereb Cortex. 2022;32(17):3763–76. pmid:34875678
View Article
PubMed/NCBI
Google Scholar
31.
de Cheveigné A, Wong DDE, Di Liberto GM, Hjortkjær J, Slaney M, Lalor E. Decoding the auditory brain with canonical component analysis. Neuroimage. 2018;172:206–16. pmid:29378317
View Article
PubMed/NCBI
Google Scholar
32.
Heilbron M, Armeni K, Schoffelen J-M, Hagoort P, de Lange FP. A hierarchy of linguistic predictions during natural language comprehension. Proc Natl Acad Sci U S A. 2022;119(32):e2201968119. pmid:35921434
View Article
PubMed/NCBI
Google Scholar
33.
Broderick MP, Anderson AJ, Lalor EC. Semantic context enhances the early auditory encoding of natural speech. J Neurosci. 2019;39(38):7564–75. pmid:31371424
View Article
PubMed/NCBI
Google Scholar
34.
O’Sullivan J, Chen Z, Herrero J, McKhann GM, Sheth SA, Mehta AD, et al. Neural decoding of attentional selection in multi-speaker environments without access to clean sources. J Neural Eng. 2017;14(5):056001. pmid:28776506
View Article
PubMed/NCBI
Google Scholar
35.
Choudhari V, Han C, Bickel S, Mehta AD, Schevon C, McKhann GM, et al. Brain-controlled augmented hearing for spatially moving conversations in multi-talker environments. Adv Sci (Weinh). 2024;11(41):e2401379. pmid:39248654
View Article
PubMed/NCBI
Google Scholar
36.
Alickovic E, Lunner T, Gustafsson F, Ljung L. A tutorial on auditory attention identification methods. Front Neurosci. 2019;13:153. pmid:30941002
View Article
PubMed/NCBI
Google Scholar
37.
Gillis M, Vanthornhout J, Simon JZ, Francart T, Brodbeck C. Neural markers of speech comprehension: measuring EEG tracking of linguistic speech representations, controlling the speech acoustics. J Neurosci. 2021;41(50):10316–29. pmid:34732519
View Article
PubMed/NCBI
Google Scholar
38.
Alickovic E, Ng EHN, Fiedler L, Santurette S, Innes-Brown H, Graversen C. Effects of hearing aid noise reduction on early and late cortical representations of competing talkers in noise. Front Neurosci. 2021;15:636060. pmid:33841081
View Article
PubMed/NCBI
Google Scholar
39.
Raghavan VS, O’Sullivan J, Herrero J, Bickel S, Mehta AD, Mesgarani N. Improving auditory attention decoding by classifying intracranial responses to glimpsed and masked acoustic events. Imaging Neurosci (Camb). 2024;2:imag–2–00148. pmid:39867597
View Article
PubMed/NCBI
Google Scholar
40.
Hjortkjær J, Wong DDE, Catania A, Märcher-Rørsted J, Ceolini E, Fuglsang SA, et al. Real-time control of a hearing instrument with EEG-based attention decoding. J Neural Eng. 2025;22(1):10.1088/1741-2552/ad867c. pmid:39996608
View Article
PubMed/NCBI
Google Scholar
41.
Geirnaert S, Vandecappelle S, Alickovic E, de Cheveigne A, Lalor E, Meyer BT, et al. Electroencephalography-based auditory attention decoding: toward neurosteered hearing devices. IEEE Signal Process Mag. 2021;38(4):89–102.
View Article
Google Scholar
42.
Wilroth J, Alickovic E, Skoglund MA, Signoret C, Rönnberg J, Enqvist M. Improving tracking of selective attention in hearing aid users: the role of noise reduction and nonlinearity compensation. eNeuro. 2025;12(2):ENEURO.0275-24.2025. pmid:39880674
View Article
PubMed/NCBI
Google Scholar
43.
Katthi JR, Ganapathy S, Kothinti S, Slaney M. Deep canonical correlation analysis for decoding the auditory brain. Annu Int Conf IEEE Eng Med Biol Soc. 2020;2020:3505–8. pmid:33018759
View Article
PubMed/NCBI
Google Scholar
44.
Fuglsang SA, Dau T, Hjortkjær J. Noise-robust cortical tracking of attended speech in real-world acoustic scenes. Neuroimage. 2017;156:435–44. pmid:28412441
View Article
PubMed/NCBI
Google Scholar
45.
Shinn-Cunningham BG. Object-based auditory and visual attention. Trends Cogn Sci. 2008;12(5):182–6. pmid:18396091
View Article
PubMed/NCBI
Google Scholar
46.
Bregman AS. Auditory scene analysis: the perceptual organization of sound. The MIT Press; 1990.
View Article
Google Scholar
47.
Corbetta M, Shulman GL. Control of goal-directed and stimulus-driven attention in the brain. Nat Rev Neurosci. 2002;3(3):201–15. pmid:11994752
View Article
PubMed/NCBI
Google Scholar
48.
Corbetta M, Patel G, Shulman GL. The reorienting system of the human brain: from environment to theory of mind. Neuron. 2008;58(3):306–24. pmid:18466742
View Article
PubMed/NCBI
Google Scholar
49.
Ding N, Simon JZ. Emergence of neural encoding of auditory objects while listening to competing speakers. Proc Natl Acad Sci U S A. 2012;109(29):11854–9. pmid:22753470
View Article
PubMed/NCBI
Google Scholar
50.
Hausfeld L, Riecke L, Valente G, Formisano E. Cortical tracking of multiple streams outside the focus of attention in naturalistic auditory scenes. Neuroimage. 2018;181:617–26. pmid:30048749
View Article
PubMed/NCBI
Google Scholar
51.
Broderick MP, Anderson AJ, Di Liberto GM, Crosse MJ, Lalor EC. Electrophysiological correlates of semantic dissimilarity reflect the comprehension of natural, narrative speech. Curr Biol. 2018;28(5):803-809.e3. pmid:29478856
View Article
PubMed/NCBI
Google Scholar
52.
Anderson AJ, Davis C, Lalor EC. Deep-learning models reveal how context and listener attention shape electrophysiological correlates of speech-to-language transformation. PLoS Comput Biol. 2024;20(11):e1012537. pmid:39527649
View Article
PubMed/NCBI
Google Scholar
53.
Brodbeck C, Jiao A, Hong LE, Simon JZ. Neural speech restoration at the cocktail party: Auditory cortex recovers masked speech of both attended and ignored speakers. PLoS Biol. 2020;18(10):e3000883. pmid:33091003
View Article
PubMed/NCBI
Google Scholar
54.
Orf M, Wöstmann M, Hannemann R, Obleser J. Target enhancement but not distractor suppression in auditory neural tracking during continuous speech. iScience. 2023;26(6):106849. pmid:37305701
View Article
PubMed/NCBI
Google Scholar
55.
Fiedler L, Wöstmann M, Herbst SK, Obleser J. Late cortical tracking of ignored speech facilitates neural selectivity in acoustically challenging conditions. Neuroimage. 2019;186:33–42. pmid:30367953
View Article
PubMed/NCBI
Google Scholar
56.
Raghavan VS, O’Sullivan J, Bickel S, Mehta AD, Mesgarani N. Distinct neural encoding of glimpsed and masked speech in multitalker situations. PLoS Biol. 2023;21(6):e3002128. pmid:37279203
View Article
PubMed/NCBI
Google Scholar
57.
Carta S, Aličković E, Zaar J, Valdés AL, Di Liberto GM. Cortical encoding of phonetic onsets of both attended and ignored speech in hearing impaired individuals. PLoS One. 2024;19(11):e0308554. pmid:39576775
View Article
PubMed/NCBI
Google Scholar
58.
Dimitrijevic A, Smith ML, Kadis DS, Moore DR. Neural indices of listening effort in noisy environments. Sci Rep. 2019;9(1):11278. pmid:31375712
View Article
PubMed/NCBI
Google Scholar
59.
Seifi Ala T, Graversen C, Wendt D, Alickovic E, Whitmer WM, Lunner T. An exploratory study of EEG alpha oscillation and pupil dilation in hearing-aid users during effortful listening to continuous speech. PLoS One. 2020;15(7):e0235782. pmid:32649733
View Article
PubMed/NCBI
Google Scholar
60.
Paul BT, Chen J, Le T, Lin V, Dimitrijevic A. Cortical alpha oscillations in cochlear implant users reflect subjective listening effort during speech-in-noise perception. PLoS One. 2021;16(7):e0254162. pmid:34242290
View Article
PubMed/NCBI
Google Scholar
61.
Fiedler L, Seifi Ala T, Graversen C, Alickovic E, Lunner T, Wendt D. Hearing aid noise reduction lowers the sustained listening effort during continuous speech in noise-a combined pupillometry and EEG study. Ear Hear. 2021;42(6):1590–601. pmid:33950865
View Article
PubMed/NCBI
Google Scholar
62.
Shahsavari Baboukani P, Graversen C, Alickovic E, Østergaard J. Speech to noise ratio improvement induces nonlinear parietal phase synchrony in hearing aid users. Front Neurosci. 2022;16:932959. pmid:36017182
View Article
PubMed/NCBI
Google Scholar
63.
Lakatos P, Barczak A, Neymotin SA, McGinnis T, Ross D, Javitt DC, et al. Global dynamics of selective attention and its lapses in primary auditory cortex. Nat Neurosci. 2016;19(12):1707–17. pmid:27618311
View Article
PubMed/NCBI
Google Scholar
64.
Wöstmann M, Lim S-J, Obleser J. The human neural alpha response to speech is a proxy of attentional control. Cereb Cortex. 2017;27(6):3307–17. pmid:28334352
View Article
PubMed/NCBI
Google Scholar
65.
Foxe JJ, Snyder AC. The role of alpha-band brain oscillations as a sensory suppression mechanism during selective attention. Front Psychol. 2011;2:154. pmid:21779269
View Article
PubMed/NCBI
Google Scholar
66.
Herrmann B, Maess B, Henry MJ, Obleser J, Johnsrude IS. Neural signatures of task-related fluctuations in auditory attention and age-related changes. Neuroimage. 2023;268:119883. pmid:36657693
View Article
PubMed/NCBI
Google Scholar
67.
McCloy DR, Lau BK, Larson E, Pratt KAI, Lee AKC. Pupillometry shows the effort of auditory attention switching. J Acoust Soc Am. 2017;141(4):2440. pmid:28464660
View Article
PubMed/NCBI
Google Scholar
68.
Tune S, Alavash M, Fiedler L, Obleser J. Neural attentional-filter mechanisms of listening success in middle-aged and older individuals. Nat Commun. 2021;12(1):4533. pmid:34312388
View Article
PubMed/NCBI
Google Scholar
69.
Decruy L, Lesenfants D, Vanthornhout J, Francart T. Top-down modulation of neural envelope tracking: the interplay with behavioral, self-report and neural measures of listening effort. Eur J Neurosci. 2020;52(5):3375–93. pmid:32306466
View Article
PubMed/NCBI
Google Scholar
70.
Polyn SM, Norman KA, Kahana MJ. A context maintenance and retrieval model of organizational processes in free recall. Psychol Rev. 2009;116(1):129–56. pmid:19159151
View Article
PubMed/NCBI
Google Scholar
71.
Baror S, Cohen M, Haik N, Avraham G, Ben-Yakov A. The role of context in continuity and segmentation. Nat Hum Behav. 2026;10(5):988–1005. pmid:41673193
View Article
PubMed/NCBI
Google Scholar
72.
Sinclair AH, Manalili GM, Brunec IK, Adcock RA, Barense MD. Prediction errors disrupt hippocampal representations and update episodic memories. Proc Natl Acad Sci U S A. 2021;118(51):e2117625118. pmid:34911768
View Article
PubMed/NCBI
Google Scholar
73.
Caucheteux C, Gramfort A, King J-R. Evidence of a predictive coding hierarchy in the human brain listening to speech. Nat Hum Behav. 2023;7(3):430–41. pmid:36864133
View Article
PubMed/NCBI
Google Scholar
74.
Mischler G, Li YA, Bickel S, Mehta AD, Mesgarani N. Contextual feature extraction hierarchies converge in large language models and the brain. Nat Mach Intell. 2024;6(12):1467–77.
View Article
Google Scholar
75.
Goldstein A, Zada Z, Buchnik E, Schain M, Price A, Aubrey B, et al. Shared computational principles for language processing in humans and deep language models. Nat Neurosci. 2022;25(3):369–80. pmid:35260860
View Article
PubMed/NCBI
Google Scholar
76.
LCM Team, Barrault L, Duquenne P-A, Elbayad M, Kozhevnikov A, Alastruey B, et al. Large concept models: Language modeling in a sentence representation space. arXiv. 2024.
View Article
Google Scholar
77.
Tikochinski R, Goldstein A, Meiri Y, Hasson U, Reichart R. Incremental accumulation of linguistic context in artificial and biological neural networks. Nat Commun. 2025;16(1):803. pmid:39824935
View Article
PubMed/NCBI
Google Scholar
78.
Schiel F. Automatic phonetic transcription of non-prompted speech. In: 1999.
79.
Schiel F. A statistical model for predicting pronunciation. 2015.
80.
Kisler T, Reichel U, Schiel F. Multilingual processing of speech via web services. Comput Speech Lang. 2017;45:326–47.
View Article
Google Scholar
81.
Boersma P. Praat: doing phonetics by computer. 2007. Available from:
http://www.praat.org/
View Article
Google Scholar
82.
Jiang AQ, Sablayrolles A, Mensch A, Bamford C, Chaplot DS, de las Casas D, et al. Mistral 7B. arXiv. 2023.
View Article
Google Scholar
83.
Chalehchaleh A, Winchester MM, Di Liberto GM. Robust assessment of the cortical encoding of word-level expectations using the temporal response function. J Neural Eng. 2025;22(1):10.1088/1741-2552/ada30a. pmid:39719122
View Article
PubMed/NCBI
Google Scholar
84.
Delorme A, Makeig S. EEGLAB: an open source toolbox for analysis of single-trial EEG dynamics including independent component analysis. J Neurosci Methods. 2004;134(1):9–21. pmid:15102499
View Article
PubMed/NCBI
Google Scholar