Where is the melody? Spontaneous attention orchestrates melody formation during polyphonic music listening

  1. Martin M Winchester
  2. Kevin Reynolds
  3. Charbel Nebo
  4. Ian Cecil Scott
  5. Giovanni M Di Liberto

  • Curated by eLife

    eLife logo

    eLife Assessment

    This valuable work potentially advances our understanding of melody extraction in polyphonic music listening by identifying spontaneous attentional focus in uninstructed listening contexts. However, the evidence supporting the main conclusions is incomplete. The work will be of interest to psychologists and neuroscientists working on music listening, attention, and perception in ecological settings.

Reviewed by

Read the full article

Listed in

Log in to save this article

Abstract

Humans seamlessly process multi-voice music into a coherent perceptual whole. Yet the neural strategies supporting this experience remain unclear. One fundamental component of this process is the formation of melody, a core structural element of music. Previous work on monophonic listening has provided strong evidence for the neurophysiological basis of melody processing, for example indicating predictive processing as a foundational mechanism underlying melody encoding. However, considerable uncertainty remains about how melodies are formed during polyphonic music listening, as existing theories (e.g., divided attention, figure–ground model, stream integration) fail to unify the full range of empirical findings. Here, we combined behavioural measures with non-invasive electroencephalography (EEG) to probe spontaneous attentional bias and melodic expectation while participants listened to two-voice classical excerpts. Our uninstructed listening paradigm eliminated a major experimental constraint, creating a more ecologically valid setting. We found that attention bias was significantly influenced by both the high-voice superiority effect and intrinsic melodic statistics. We then employed transformer-based models to generate next-note expectation profiles and test competing theories of polyphonic perception. Drawing on our findings, we propose a weighted-integration framework in which attentional bias calibrates the overall degree of integration of the competing streams. In doing so, the proposed framework reconciles previous divergent accounts by showing that, even under free-listening conditions, melodies emerge through an attention-guided statistical integration mechanism.

Article activity feed

  1. eLife

    eLife Assessment

    This valuable work potentially advances our understanding of melody extraction in polyphonic music listening by identifying spontaneous attentional focus in uninstructed listening contexts. However, the evidence supporting the main conclusions is incomplete. The work will be of interest to psychologists and neuroscientists working on music listening, attention, and perception in ecological settings.

  2. eLife

    Reviewer #1 (Public review):

    Summary:

    This manuscript investigates the interplay between spontaneous attention and melody formation during polyphonic music listening. The authors use EEG recordings during uninstructed listening to examine how attention bias influences melody processing, employing both behavioural measures and computational modelling with music transformers. The study introduces a very clever pitch-inversion manipulation design to dissociate high-voice superiority from melodic salience and propose a "weighted integration" model where attention dynamically modulates how multiple voices are combined into perceived melody.

    Strengths:

    (1) The attention bias findings (Figure 2) are compelling and methodologically sound, with convergent evidence from both behavioral and neural measures.

    (2) The pitch-inversion manipulation appears to super elegantly dissociate two competing factors (high-voice superiority vs melodic salience), moreover, the authors claim that the chosen music lends itself perfectly to his PolyInv condition. A claim I cannot really evaluate but which would make it even more neat.

    (3) Nice linking between hypotheses and operationalisations.

    Weaknesses:

    My main remaining concerns are about Figure 3, which appears central (e.g. the title) and where the evidence remains, I'm afraid, incomplete.

    (1) Statistics. As R3 observed, the stats in Fig 3 are across repetitions and not over subjects -- therefore the p-value does not appear to support population inference, and the key striking interaction in the figure does not appear to be tested? I don't see how the author response addresses this.

    (2) Model checks. I asked how correlated the monophonic and polyphonic surprisal values are with each other. The revision reports correlations before and after inversion, which answers a different question. From Figure 3c it appears to be the case that the poly feature adds nothing to classification in PolyOrig, so it is also unclear whether Figure 3 reflects attention or a mis-specified feature.

    I also noticed that the statistics in Figure 3 changed (see recommendation)

  3. eLife

    Reviewer #2 (Public review):

    Summary:

    The authors sought to understand the drivers of spontaneous attentional bias and melodic expectation generation during listening to short two-part classical pieces. They measured scalp EEG data in a monophonic condition and trained a model to reconstruct the audio envelope from the EEG. They then used this model to probe which of the two voices was best reflected in the neural signal during two polyphonic conditions. In one condition the original piece was presented, in the other the voices were switched in an attempt to distinguish between effects of (a) the pitch range of one voice compared to the other and (b) intrinsic melodic features. They also collected a behavioural measure of attentional bias for a subset of the stimuli in a separate study. Further modelling assessed whether expectations of how the melody would unfold were formed based on an integrated percept of melody across the two voices or based on a single voice. The authors sought to relate the findings to different theories of how musical/auditory scene analysis occurs, based on divided attention, figure-ground perception, and stream integration.

    Strengths:

    (1) A clever but simple manipulation - transposing the voices such that the higher one became the lower one - allowed an assessment of different factors that might affect the allocation of attention.

    (2) State-of-the-art analytic techniques were applied to (a) build a music attention decoder (these are more commonly encountered for speech) (b) relate the neural data to features of the stimulus at the level of acoustics and expectation.

    (3) The effects appeared robust across the group, not driven by a handful of participants.

    Weaknesses:

    (1) Although melody statistics are mentioned throughout, none have been reported. It would be helpful to see the features that lead to "motif attractiveness" quantified, as well as how they differ across lines. The work of David Huron such as at https://dl.acm.org/doi/abs/10.1145/3469013.3469016 provides examples that could be calculated with ease and compared across the two lines: "the tendency for small over large pitch movements, for large leaps to ascend, for musical phrases to fall in pitch, and for phrases to begin with an initial pitch rise". The authors also mention differences in ornamentation. Such comparisons would make more tangible for the reader what does differ across the original "melody" and "support" line. In particular, as the authors themselves note, lines in double-counterpoint pieces can to a degree operate interchangeably. Bach's inventions in particular use a lot of direct repetition (up to octave invariance), which one would expect to minimise differences in the statistics mentioned. If there were differences in the number of note onsets in the "melody" and "support" line, the implications for attentional capture and biasing of TRF estimates need to be considered. references purporting to relate to melodic statistics (11-14 in original numbering) seem rather to relate to high-voice superiority.

    (2) Clearer details of how the integration model was implemented will be critical to interpreting the results relating to melodic expectations. It is not obvious how notes with simultaneous onsets were included in the model.

    (3) The behavioural and neural measures of attentional bias are time-averaged over a stimulus, but readers may be left with the impression that they were time-resolved, given references to "integration rate" (for example).

  4. eLife

    Reviewer #3 (Public review):

    Summary:

    In this paper, Winchester and colleagues investigated melodic perception in natural music listening. They highlight the central role of attentional processes in identifying one particular stream in polyphonic material, and propose to compare several theoretical accounts, namely (1) divided attention, (2) figure-ground separation and (3) stream integration. In parallel, the authors compare the relative strength of exogenous attentional effects (i.e. salience) produced by two common traits of melodies: high-pitch (compared to other voices), and attractive statistics. To ensure the generalisability of their results to real-life listening contexts, they developed a new uninstructed listening paradigm in which participants can freely attend to any part of a musical stimulus.

    Major strengths and weaknesses of the methods and results:

    (1) Winchester and colleagues capitalized on previous attention decoding techniques and proposed an uninstructed listening paradigm, used here to investigate the spontaneous attentional focus during listening. The EEG decoding results are coherent with the behavioral data, suggesting that the paradigm is robust and relevant.

    The authors first evaluate the relative importance of high-pitch and statistics in producing an attentional bias (Fig. 2). Behavioral results show a clear pattern, in which both effects are present with a dominance of the high-pitch one. The only weakness inherent to this protocol, which the authors acknowledge, is that behavioral responses are measured on a second presentation of short samples, which may induce different attentional focus than in the first uninstructed listening.
    The EEG analyses then compare the decoding of each melody (the high or low voice, with "richer" or "poorer" statistics) and show a similar pattern. This leaves open a confound factor: the decoder is trained on the monophonic condition, so any difference in how well the four melodies are decoded, prior to any attentional effect, would also be expected in the polyphonic conditions. The authors report in their response that the decoder performs equally well on the four monophonic conditions. This addresses the right question, but it concludes from a non-significant result, and the tests provided are unpaired although the design is within-participant.

    This control does not appear in the manuscript, which states (l.128-130) that the decoders are "unbiased with regard to pitch height and voice statistics" without showing the data that would support this. Moreover, to capture attentional biases specifically, the decoding scores in the polyphonic conditions should be compared to the scores in the monophonic conditions, and attention could be expected to increase the decoding of the attended stream or decrease the unattended one.

    (2) Winchester and colleagues then evaluate the encoding of melodic surprise and uncertainty (Fig. 3), comparing a monophonic and a polyphonic model (Anticipatory Music Transformer) with a CCA analysis. In Fig. 3 panel C, the statistics are not valid. They are computed across the 250 repetitions of the match-vs-mismatch procedure and not across the 31 participants: every ANOVA is reported as F(1,249). These statistics therefore do not support conclusions at the population level (the variability between participants never enters the error term), and they depend on the number of repetitions, a parameter of the analysis rather than a property of the data. The statistics reported for Figure 3 also differ from those reported for the same, unchanged figure in the previous version, and the change is not documented. The condition * expectations interaction, for instance, is now significant where it was not. The CCA analysis also cannot be reproduced from the public repository (osf.io/bjdh6), which contains no code for the CCA or the match-vs-mismatch classification. The stimuli are absent as well. This is not only a formal issue: the repository defines many alternative feature sets, and the acoustic set named in the code does not correspond to the one described in the Methods, so a reader cannot determine which features produced Figure 3.

    Finally, the finding of Fig. 3C is not investigated further: a comparison of the encoding of individual melodies (e.g. AMmono high-voice vs AMmono low-voice, in PolyOrig and PolyInv conditions) is not provided, and no subsequent analysis or experiment accounts for the double dissociation. The result appears more like a starting point for further exploration than a finding that can support theoretical claims.

    An appraisal of whether the authors achieved their aims, and whether the results support their conclusions:

    (1) Winchester and colleagues aimed at identifying the melodic stream that attracts attention during the listening of natural polyphonic music, and the underlying attentional processes. Their behavioral results confirm that high-pitch and attractive statistics increase melodic salience with a greater effect size of the former. Their interpretation of the CCA results, in terms of stream segregation under high melodic salience and stream integration under weaker attentional biases, seems supported by the data, but the three limitations above (statistics, reproducibility, absence of explanatory analyses) prevent it from being evaluated. The relevance of the Anticipatory Music Transformer as a model of the listeners' expectations, compared with a well-established model such as IDyOM, is not discussed. Of note, the melodic statistics reported in the authors' response show small and mostly non-significant differences between the two voices, which weakens the motif attractiveness account.

    (2) Overall, the authors achieved most of the aims presented in the introduction, although they couldn't give a more precise account on the attentional processes at stake. Their interpretations are sound and not overstated.

    Impact of the work on the field, and the utility of the methods and data to the community:

    The uninstructed listening paradigm introduced in this paper will likely have an impact for psychologists and neuroscientists working on music perception and auditory attention, enabling them to conduct experiments in more ecological settings. While the attentional biases towards melodies with high-pitch and attractive statistics are already known, showing their relative effect is an important step in building precise models of auditory attention. The release of the EEG data and stimulus features in a standardized format is a useful contribution in itself and would be considerably more so with the analysis code and the stimuli. Future work may be necessary to identify the models (markov-chains, deep-learning) and setup (data analysis, stimuli, control variables) that do or do not replicate these results.

  5. eLife

    Author response:

    The following is the authors’ response to the original reviews.

    Reviewer #1 (Public review):

    Summary:

    This manuscript investigates the interplay between spontaneous attention and melody formation during polyphonic music listening. The authors use EEG recordings during uninstructed listening to examine how attention bias influences melody processing, employing both behavioural measures and computational modelling with music transformers. The study introduces a very clever pitch-inversion manipulation design to dissociate high-voice superiority from melodic salience, and proposes a "weighted integration" model where attention dynamically modulates how multiple voices are combined into perceived melody.

    Strengths:

    (1) The attention bias findings (Figure 2) are compelling and methodologically sound, with convergent evidence from both behavioral and neural measures.

    (2) The pitch-inversion manipulation appears to super elegantly dissociate two competing factors (high-voice superiority vs melodic salience), moreover, the authors claim that the chosen music lends itself perfectly to his PolyInv condition. A claim I cannot really evaluate, but which would make it even more neat.

    (3) Nice bridge between hypotheses and operationalisations.

    Weaknesses:

    The results in Figure 3 are very striking, but I have a number of questions before I can consider myself convinced.

    (1) Conceptual questions about surprisal analysis:

    The pattern of results seems backwards to me. Since the music is inherently polyphonic in PolyOrig, I'd expect the polyphonic model to fit the brain data better - after all, that's what the music actually is. These voices were composed to interact harmonically, so modeling them as independent monophonic streams seems like a misspecification. Why would the brain match this misspecified model better?

    Conversely, it would seem to me the pitch inversion in PolyInv disrupts (at least to some extent) the harmonic coherence, so if anywhere, I'd a priori expect that in this condition, listeners would rather be processing streams separately - making the monophonic model fit better there (or less bad), not in PolyOrig. The current pattern is exactly opposite to what seems logical to me.

    The reviewer raised an important point that we have further clarified in the text (lines 503-516). We agree that a physiologically valid computational model of polyphonic music processing should match neural signals more closely than a monophonic model for all conditions. However, these transformer models are not built to be physiologically plausible. Their only objective is to minimise the next-event prediction error based on the prior context, which may also be achieved in ways that differ from human brain processing.

    Previous studies (Di Liberto et al., eLife, 2020; Marion et al., Journal of Neuroscience, 2021; Kern et al., eLife, 2022), combined with our results on the monophonic condition, makes us confident that the monophonic models are correlated with how our brains process melodies from monophonic streams. Our finding that monophonic models also correlate with neural activity during PolyOrig is quite exciting, and far from obvious. Finding that melodic surprises derived from a polyphonic model work outperform the monophonic model at all is also extremely interesting. In fact, that result suggests that, under certain conditions (in this case, corresponding to a change in attention bias), our brains build melodic surprises by integrating information from multiple streams. Indeed, it is important to thoroughly discuss the motivation, reasoning, and caveats of this result, and we have done our best to augment the manuscript in that regard.

    Our rationale was the following: In the PolyOrig condition, there is a strong attention bias, meaning that the melody corresponds primarily to one specific stream, and melodic expectations seem to be built on that stream exclusively (i.e., monophonic model). One caveat here is that our polyphonic model may simply be a “bad” model (i.e., non-physiologically plausible) for PolyOrig. Nonetheless, we find the result extremely promising, as the monophonic model is a good predictor of the EEG signal, and that its classification gain (AM-A) was comparable between PolyOrig and Monophonic condition.

    We think that more work is needed to identify a polyphonic model that is closest to brain neural activity. That this question should be investigated using multiple datasets with various stimuli and conditions, and we are doing our best to provide a wider range of data and conditions in our ongoing work (for example, we are currently collecting data with three-stream music).

    (2) Missing computational analyses:

    If the transformer is properly trained, it should "understand" (i.e., predict/compress) the polyphonic music better, right? Can the authors demonstrate this via perplexity scores, bitsper-byte, or other prediction metrics, comparing how well each model (polyphonic vs monophonic) handles the music in both conditions? Similarly, if PolyInv truly maintains musical integrity as claimed, the polyphonic model should handle it as well as PolyOrig. But if the inversion does disrupt the music, we should see this reflected in degraded prediction scores. These metrics would validate whether the experimental manipulation works as intended. Also, how strongly are the surprisal streams correlated? There are many non-trivial modelling steps that should be reported in more detail.

    The revised manuscript includes an assessment of the impact of the inversion on the model predictions. As expected, the inversion alters the ability of the model to predict the next note, producing a statistically significant increase in the surprise (KS test p = 9.7*10-130; D = 0.116), entropy (KS test p = 6.9*10-15; D = 0.0388), and perplexity (Wilcoxon signed-rank test p = 9.4*10-6, r = -0.783) for the polyphonic model. While our manipulation led to a reduced predictive ability of the model, it should be noted that these are very small effects (D<0.2).

    Next, we measured the Pearson’s correlations of the expectation values for corresponding notes before and after the inversion of the polyphonic tracks, finding relatively low correlation values for surprise (r = 0.150, p = 7*10-68) and entropy (r = 0.106, p = 3*10-34).

    To test whether the low correlations between the original and inverted expectation streams affected the neural results, we re-ran the CCA analysis used in Figure 3 on the PolyInv condition using expectation features extracted from PolyOrig. This control removes the inversion-induced change in the expectation features. The pattern of results was unchanged, indicating that the main findings are robust to these differences in model-derived expectation values. The revised manuscript includes this new data and a reflection on the implication of these results (lines 348-370; please see the new Figure 3 Supplement 1).

    (3) Methodological inconsistencies:

    Why are the two main questions (Figures 2 and 3) answered with completely different analytical approaches? The switch from TRF to CCA with match-vs-mismatch classification seems unmotivated. I think it's very important to provide a simpler model comparison - just TRF with acoustic features plus either polyphonic or monophonic surprisal - evaluated on relevant electrodes or the full scalp. This would make the results more comparable and interpretable.

    The revised manuscript includes an explanation on that choice (lines 309-315). TRFs are effective when studying selective attention, which has been widely documented in the speech selective attention literature. The TRF also leads to interpretable results (ERP-like), which is preferable. However, when the objective is to assess the neural encoding of a certain property, CCA returns stronger scores as it accounts for all stimulus features and all EEG channels simultaneously. This is preferable in this case, as the EEG signatures of melodic expectation typically have a small magnitude (r~0.005; see Figure 2C in Di Liberto et al., eLife, 2020). Interpretability (i.e., the ability to infer spatial and temporal patterns of the neural response) was also less important on this analysis, as the two models were expected to simply be better or worse representations of the same neural sources. As such, we opted for the method with the highest sensitivity in terms of explained variance (here operationalised as a classification metric - as explored in our previous study on phonotactics, Di Liberto et al., NeuroImage, 2019).

    (4) Presentation and methods:

    (a) Coming from outside music/music theory, I found the paper somewhat abstract and hard to parse initially. The experimental logic becomes clearer with reflection, but you're doing yourselves a disservice with the jargon-heavy presentation. It would be useful to include example stimuli.

    We have added an example sound file, with the corresponding PolyOrig, PolyInv versions (Figure 1 - supplement 1).

    With regard to the wording in the manuscript, we agree that it may sound heavy in some parts. Nonetheless, that was an inevitable consequence of avoiding confusion and ambiguity. We are referring, for example, to the PolyInv manipulation, where it was critical to use a consistent wording. For instance, high-pitch stream could be intended as the stream that has the higher pitch in PolyInv, or the stream that had originally the higher pitch in PolyOrig. This was an issue of an early version of the manuscript that was fixed, regrettable making the manuscript heavier to read. We remain open to suggestions and did our best to improve some of our wording in the revised version (especially in the Discussion section).

    (b) The methods section is extremely brief - no details whatsoever are provided regarding the modelling: What specific music transformer architecture? Which implementation of this "anticipatory music transformer"? Pre-trained on what corpus - monophonic, polyphonic, Western classical only? What constituted "technical issues" for the 9 excluded participants? What were the channel rejection criteria?

    The methods section has been substantially augmented to include all details necessary for replication of our analyses (see lines 578-580, 670-737).

    Reviewer #2 (Public review):

    Summary:

    The authors sought to understand the drivers of spontaneous attentional bias and melodic expectation generation during listening to short two-part classical pieces. They measured scalp EEG data in a monophonic condition and trained a model to reconstruct the audio envelope from the EEG. They then used this model to probe which of the two voices was best reflected in the neural signal during two polyphonic conditions. In one condition, the original piece was presented, in the other, the voices were switched in an attempt to distinguish between effects of (a) the pitch range of one voice compared to the other and (b) intrinsic melodic features. They also collected a behavioural measure of attentional bias for a subset of the stimuli in a separate study. Further modelling assessed whether expectations of how the melody would unfold were formed based on an integrated percept of melody across the two voices, or based on a single voice. The authors sought to relate the findings to different theories of how musical/auditory scene analysis occurs, based on divided attention, figure-ground perception, and stream integration.

    Strengths:

    (1) A clever but simple manipulation - transposing the voices such that the higher one became the lower one - allowed an assessment of different factors that might affect the allocation of attention.

    (2) State-of-the-art analytic techniques were applied to (a) build a music attention decoder (these are more commonly encountered for speech) and (b) relate the neural data to features of the stimulus at the level of acoustics and expectation.

    (3) The effects appeared robust across the group, not driven by a handful of participants.

    Weaknesses:

    (1) A key goal of the work is to establish the relative importance for the listener's attention of a voice's (a) mean pitch in the context of the two voices (high-voice superiority) and (b) intrinsic melodic statistics/motif attractiveness. The rationale of the experimental manipulation is that switching the relative height of the lines allows these to be dissociated by imparting the same high-voice benefit to the new high-voice and the same preferred intrinsic melodic statistics to the new low voice. However, previous work suggests that the high-voice superiority effect is not all-or-nothing. Electrophysiology supported by auditory nerve modelling found it to depend on the degree of voice separation in a non-monotonic way (see https://doi.org/10.1016/j.heares.2013.07.014 at p. 68). Although the authors keep the overall pitch of the lower (and upper) line fixed across conditions, systematically different contour patterns across the voices could give rise to a sub-optimal distribution of separations in the PolyInv versus PolyOrig condition. This could weaken the high-voice superiority effect in PolyInv and explain the pattern of results. One could argue that such contour differences are examples of the "intrinsic melodic statistics" put forward as the effect working in opposition to high-voice superiority, but it is their interaction across voices that matters here.

    We thank the reviewer for this thoughtful comment. We agree that high‑voice superiority can vary with between‑stream pitch separation and need not be all‑or‑nothing. To address this, we quantified the separation differences between PolyOrig and PolyInv. Separation was higher in PolyInv (paired t‑test: p = 3.6×10⁻⁶). If a larger separation weakened the high-voice superiority effect, this would threaten our interpretation of the result in Figure 2 B, D (where PolyOrig presented a stronger high-voice attention bias than PolyInv). However, pitch separation did not correlate with our EEG attention‑bias index when considering both conditions together (r = 0.005, p = 0.908) or PolyOrig alone (r = -0.101, p = 0.0693). Within PolyInv, the correlation was significant and positive (r = 0.154, p = 0.00561) i.e., greater separation was associated with a stronger, not weaker, high‑voice bias. Thus, the observed results are not explained by a sub‑optimal distribution of separations across conditions. While contour differences could be viewed as part of the “intrinsic melodic statistics,” the cross‑voice interaction via separation does not account for the pattern we report.

    (2) Although melody statistics are mentioned throughout, none have been calculated. It would be helpful to see the features that presumably lead to "motif attractiveness" quantified, as well as how they differ across lines. The work of David Huron, such as at https://dl.acm.org/doi/abs/10.1145/3469013.3469016, provides examples that could be calculated with ease and compared across the two lines: "the tendency for small over large pitch movements, for large leaps to ascend, for musical phrases to fall in pitch, and for phrases to begin with an initial pitch rise". The authors also mention differences in ornamentation. Such comparisons would make it more tangible for the reader as to what differs across the original "melody" and "support" line. In particular, as the authors themselves note, lines in double-counterpoint pieces can, to a degree, operate interchangeably. Bach's inventions in particular use a lot of direct repetition (up to octave invariance), which one would expect to minimise differences in the statistics mentioned. The references purporting to relate to melodic statistics (11-14 in original numbering) seem rather to relate to high-voice superiority.

    We agree that quantifying “motif attractiveness” will be useful to the readers. We computed four standard melodic-statistics measures (H: high-pitch; L: low-pitch): Pitch Proximity (mean interval size; H < L, p = 0.0747), Leap Ascension ratio (H > L, p = 0.00111), Pitch Declination (H < L, p = 0.73), and Initial Anacrusis (H > L, p = 0.422). We also note that the H line contains ~15.2% more notes than L. Taken together, we conclude that there are small albeit measurable differences between H and L streams that predict behaviour i.e., there was an attention bias toward the H voice (with behavioural and EEG metrics consistent with this interpretation). Indeed, it should be noted that these differences were relatively small, in contrast with the large contrasts that are typical of many contemporary Western pieces. This is consistent with the double-counterpoint repertoire, where lines are intentionally written to be highly interchangeable. Here, modest statistical asymmetries nonetheless track perceptual salience. To facilitate further work, we will release code, per-piece feature tables, and note-count summaries in the repository.

    (3) The exact nature of the transposition manipulation is obscured by a confusing Figure 1B, which shows an example in which the transposed line does not keep the same note-to-note interval structure as the original line.

    Thank you for pointing that out. We have corrected that panel.

    (4) The transformer model is barely described in the main text. Even readers who are familiar with the Hidden Markov Models (e.g., in IDyOM) previously used by some of the authors to model melodic surprise and entropy would benefit from a brief description in the main text at least of how transformer models are different. The Methods section goes a little further but does not mention what the training set was, nor the relative weight given to long- and short-term memory models.

    (5) The match-mismatch procedure should be explained in enough detail for readers to at least understand what value represents chance performance and why performance would be measured as an average over participants. Relatedly, there is no description at all of CCA or the match-mismatch procedure in the Methods.

    The revised manuscript includes substantial additions to the methodology description, including details on the Transformer model and the match-mismatch procedures (lines 670-716). Please also see our answer to Reviewer 1, Point 4.

    (6) Details of how the integration model was implemented will be critical to interpreting the results relating to melodic expectations. It is not clear how "a single melody combining the two streams" was modelled, given that at least some notes presumably overlapped in time.

    The revised manuscript includes details on the integration model (see lines 706-716, 722-737).

    The AMT treats the polyphonic input as a single event sequence, rather than as two separate streams, where simultaneous notes are encoded as distinct events with the same onset times. Thus, the model generated an expectation value for each note event in that sequence, including notes with identical onset times. For simultaneous notes, these expectation values are not equivalent: the first note at a given onset is evaluated without the other simultaneous note yet having been incorporated into the context, whereas the final note at that same onset is evaluated after the other simultaneous note has been processed. AMT is a pretrained model that was trained using the same representation. To ensure that the integrated expectation feature reflected the full polyphonic context present at that timepoint, we used the expectation value of the last note event at each onset time. By using this merged representation, the model captures the polyphonic context. The revised Methods now makes this explicit.

    (7) The authors propose a weighted integration model, referring in the Discussion to dynamics and an integration rate. They do show that in the PolyOrig case, the top stream bias is highest and the monophonic model gives the best prediction, while in the PolyInv case, the top stream bias is weaker and the polyphonic model provides the best prediction. However, that doesn't seem to say anything about the temporal rate of integration, just the degree, which could be fixed over the whole stimulus. Relatedly, the terms "strong attention bias" and "weak attention bias" in Highlight 4 might give the impression of different attention modes for a given listener, or perhaps different types of listeners, but this seems to be shorthand for how attention is allocated for different types of stimuli (namely those that have or have not had their voices reversed).

    Our wording was misleading and was improved in this revision (lines 500-516). Our use of “dynamic”, in that sentence, did not refer to temporal integration, but it was referring to the possibility that our integration strategy may adapt to the stimulus (as suggested by the different results in PolyOrig and PolyInv). The revised manuscript includes a clearer wording that we think solves the issue. With regard to temporal dynamics, the reviewer is absolutely correct in that our analyses have not explored the temporal rate of integration. We are currently looking into that with a more tailored experiment involving more varied stimuli, but that will be part of a separate future study.

    With regard to strong vs. weak attention bias in Highlight 4, our view is that the attention bias, in fact, leads to different “modes”, as suggested by the result in Figure 3C. Nonetheless, we appreciate that it would be useful to referring more closely to the facts (as opposed to our interpretations). As such, we have reworded Highlight 4.

    (8) Another aspect of the presentation relating to temporal dynamics is that in places (e.g., Highlight 1), the authors suggest they are tracking attention dynamically. However, as acknowledged in the Discussion, neither the behavioural nor neural measure of attentional bias are temporally resolved. The measures indicate that on average participants attend more to the higher line (less so when it formed the lower line in the original composition).

    The reviewer is correct. Our phrasing was referring to a previous version of the manuscript, as we had attempted time-resolved analyses that, however, showed quite a bit of variability, and were not necessary for the conclusions of the present study. As such, those results were excluded from our submission, as we aim to investigate the temporal dynamics aspect in a future study, including more data and more tailored denoising methodologies. We have revised the wording in the manuscript to solve this issue (please see lines 33, 39-43, 64, 262, 507).

    (9) It is not clear whether the sung-back data were analysed (and if not why participants were asked to sing the melody back rather than just listen to the two components and report which they thought was the melody). It is also not stated whether the order in which the high and low voices were played back was randomised. If not, response biases or memory capacity might have affected the behavioural attention data.

    We have added our explanation and motivations to the manuscript (lines 598-608).

    The procedure was the following: Participants were presented with a short segment of a polyphonic composition; they sung back the melody; they were presented with the two monophonic streams, one after the other; they selected which one they considered the melody.

    The numerical results from the last step of this process were analysed, while the “sung back” data was not analysed. The request of singing back before hearing the individual streams was to encourage them to answer the question based on their own melody extraction process, as opposed to perform a post hoc “reasoned” decision (possibly influenced by a presentation bias). Future research should explore whether this approach is more or less effective. Here, behavioural and EEG attention bias results are compatible, which is satisfactory for claims.

    Please note that the presentation order of the high and low-pitched melodies played was indeed randomised.

    Reviewer #3 (Public review):

    Summary:

    In this paper, Winchester and colleagues investigated melodic perception in natural music listening. They highlight the central role of attentional processes in identifying one particular stream in polyphonic material, and propose to compare several theoretical accounts, namely (1) divided attention, (2) figure-ground separation, and (3) stream integration. In parallel, the authors compare the relative strength of exogenous attentional effects (i.e., salience) produced by two common traits of melodies: high-pitch (compared to other voices), and attractive statistics. To ensure the generalisability of their results to real-life listening contexts, they developed a new uninstructed listening paradigm in which participants can freely attend to any part of a musical stimulus.

    Major strengths and weaknesses of the methods and results:

    (1) Winchester and colleagues capitalized on previous attention decoding techniques and proposed an uninstructed listening paradigm. This is an important innovation for the study of music perception in ecological settings, and it is used here to investigate the spontaneous attentional focus during listening. The EEG decoding results obtained are coherent with the behavioral data, suggesting that the paradigm is robust and relevant.

    (2) The authors first evaluate the relative importance of high-pitch and statistics in producing an attentional bias (Figure 2). Behavioral results show a clear pattern, in which both effects are present, with a dominance of the high-pitch one. The only weakness inherent to this protocol is that behavioral responses are measured based on a second presentation of short samples, which may induce a different attentional focus than in the first uninstructed listening.

    That’s a good point. That is an experimental limitation. Asking participants to repeat the melody during the first presentation (so, providing a task) could have altered the way they processed the music, so we preferred to avoid that. We instead opted for the presentation of a short snippet of each of the polyphonic pieces heard only after the initial uninstructed listening task. In principle, repeating the same identical stream may lead to different attentional patterns, we agree. We think that is less likely in our case, as we only presented very brief snippets from a random portion of each previous segment. We think that this minimises that risk, and the results are in line with that hypothesis. We have added some considerations on that risk. We actually think that the question on repetition, both from attention and melodic processing/surprise perspectives, is a very interesting one that should be investigated in dedicated studies. As such, we have added a suggestion for future work too (line 452-455).

    (3) Then, the analyses of EEG data compare the decoding results of each melody (the high or low voice, and with "richer" or "poorer" statistics), and show a similar pattern of results. However, this report leaves open the possibility of a confounding factor. In this analysis, a TRF decoding model is first trained based on the presentation of monophonic samples, and it is later used to decode the envelope of the corresponding melodies in the polyphonic scenario. The fitting scores of the training phase are not reported. If the high-pitch or richer melodies were to produce higher decoding scores during monophonic listening (due to properties of the physiological response, or to perceptual processes), a similar difference could be expected during polyphonic listening. To capture attentional biases specifically, the decoding scores in the polyphonic conditions should be compared to the scores in the monophonic conditions, and attention could be expected to increase the decoding of the attended stream or decrease the unattended one.

    We carried out additional analyses to confirm that the model was not biased toward any particular condition. The attention decoding model, which was trained on all monophonic pieces together, was evaluated on each of the four monophonic conditions (main stream high-pitch, support stream low-pitch, main stream low-pitch, support stream high-pitch). No statistically significant differences were detected when comparing the four conditions (Kruskal-Wallis = 0.377), nor when comparing high versus low pitch conditions (unpaired t-test, p= 0.567) and main versus support streams (unpaired t-test, p = 0.377).

    (4) Then, Winchester and colleagues investigate the processing of melodic information by evaluating the encoding of melodic surprise and uncertainty (Figure 3). They compare the surprise and uncertainty estimated from a monophonic or a polyphonic model (Anticipatory Music Transformer), and analyse the data with a CCA analysis. The results show a double dissociation, where the processing of melodies with a strong attentional bias (high-pitch, rich statistics) is better approximated with a monophonic model, while a polyphonic model better classifies the other melodies. While this global result is compelling, it remains a preliminary and intriguing finding, and the manuscript does not further investigate it. As it stands, the result appears more like a starting point for further exploration than a definitive finding that can support strong theoretical claims. First, it could be complemented by a comparison of the encoding of individual melodies (e.g., AMmono high-voice vs AMmono low-voice, in PolyOrig and PolyInv conditions) to highlight a more direct correspondence with the previous results (Figure 2) and allow a more precise interpretation. Second, additional analyses or experiments would be needed to unpack this result and provide greater explanatory power. Additionally, the CCA analysis is not described in the method. The statistical testing conducted on this analysis seems to be performed across the 250 repetitions of the evaluation rather than across the 40 participants, which may bias the resulting p-values. Moreover, the choice and working principle of the Anticipatory Music Transformer are not described in the method. Overall, these results seem at first glance solid, but the missing parts of the method do not allow for full evaluation or replication of them.

    For the missing methodological details please see our answer to Reviewer 1-Weakness 4, and Reviewer 2-Weakness 5, and the revised manuscript (lines 670-716, 734-737). We agree with the reviewer that our results should be taken as a stepping stone toward a better understanding of polyphonic music, and that additional work is required to draw stronger theoretical claims. We have expanded the future works section of the manuscript to reflect this consideration (lines 446-456, 500-516). The revised manuscript also clarifies that the statistical analyses on the CCA results are tests on the mean across participants, where distributions are derived by repeating the procedure 250 times for each participant (lines 324-326).

    An appraisal of whether the authors achieved their aims, and whether the results support their conclusions:

    (1) Winchester and colleagues aimed at identifying the melodic stream that attracts attention during the listening of natural polyphonic music, and the underlying attentional processes. Their behavioral results confirm that high-pitched and attractive statistics increase melodic salience with a greater effect size of the former, as stated in the discussion. The TRF analyses of EEG data seem to show a similar pattern, but could also be explained by confounding factors. Next, the authors interpret the CCA results as the results of stream segregation when there is a high melodic salience, and stream integration when there are weaker attentional biases. These interpretations seem to be supported by the data, but unfortunately, no additional analyses or experiments have been conducted to further evaluate this hypothesis. The authors also acknowledge that their results do not show whether stream segregation occurs via divided attention or figure-ground separation. However, the lack of information about the music model used (Anticipatory Music Model) and the way it was set up raises some questions about its relevance and limits as a model of cognition (e.g. Is this transformer a "better" model of the listeners' expectations than the wellestablished IDyOM model, and why?), and about the validity of those results.

    (2) Overall, the authors achieved most of the aims presented in the introduction, although they couldn't give a more precise account of the attentional processes at stake. The interpretations are sound and not overstated, with the exception of potential confounding factors that could compromise the conclusions on the neural tracking of salient melodies (EEG results, Figure 2).

    Impact of the work on the field, and the utility of the methods and data to the community: The new uninstructed listening paradigm introduced in this paper will likely have an important impact on psychologists and neuroscientists working on music perception and auditory attention, enabling them to conduct experiments in more ecological settings. While the attentional biases towards melodies with high-pitch and attractive statistics are already known, showing their relative effect is an important step in building precise models of auditory attention, and allows future paradigms to explore more fine-grained effects. Finally, the stream segregation and integration shown with this paradigm could be important for researchers working on music perception. Future work may be necessary to identify the models (Markov chains, deep learning) and setup (data analysis, stimuli, control variables) that do or do not replicate these results.

    We thank the reviewer for their work on this manuscript!

  6. Version published to 10.7554/elife.108767.2 on eLife
  7. Version published to 10.7554/elife.108767 on eLife
  8. Version published to 10.1101/2025.08.26.672294v2 on bioRxiv
  9. Version published to 10.7554/elife.108767.1 on eLife
  10. eLife

    eLife Assessment

    This valuable work potentially advances our understanding of melody extraction in polyphonic music listening by identifying spontaneous attentional focus in uninstructed listening contexts. However, the evidence supporting the main conclusions is incomplete. The work will be of interest to psychologists and neuroscientists working on music listening, attention, and perception in ecological settings.

  11. eLife

    Reviewer #1 (Public review):

    Summary:

    This manuscript investigates the interplay between spontaneous attention and melody formation during polyphonic music listening. The authors use EEG recordings during uninstructed listening to examine how attention bias influences melody processing, employing both behavioural measures and computational modelling with music transformers. The study introduces a very clever pitch-inversion manipulation design to dissociate high-voice superiority from melodic salience, and proposes a "weighted integration" model where attention dynamically modulates how multiple voices are combined into perceived melody.

    Strengths:

    (1) The attention bias findings (Figure 2) are compelling and methodologically sound, with convergent evidence from both behavioral and neural measures.

    (2) The pitch-inversion manipulation appears to super elegantly dissociate two competing factors (high-voice superiority vs melodic salience), moreover, the authors claim that the chosen music lends itself perfectly to his PolyInv condition. A claim I cannot really evaluate, but which would make it even more neat.

    (3) Nice bridge between hypotheses and operationalisations.

    Weaknesses:

    The results in Figure 3 are very striking, but I have a number of questions before I can consider myself convinced.

    (1) Conceptual questions about surprisal analysis:

    The pattern of results seems backwards to me. Since the music is inherently polyphonic in PolyOrig, I'd expect the polyphonic model to fit the brain data better - after all, that's what the music actually is. These voices were composed to interact harmonically, so modeling them as independent monophonic streams seems like a misspecification. Why would the brain match this misspecified model better?

    Conversely, it would seem to me the pitch inversion in PolyInv disrupts (at least to some extent) the harmonic coherence, so if anywhere, I'd a priori expect that in this condition, listeners would rather be processing streams separately - making the monophonic model fit better there (or less bad), not in PolyOrig. The current pattern is exactly opposite to what seems logical to me.

    (2) Missing computational analyses:

    If the transformer is properly trained, it should "understand" (i.e., predict/compress) the polyphonic music better, right? Can the authors demonstrate this via perplexity scores, bits-per-byte, or other prediction metrics, comparing how well each model (polyphonic vs monophonic) handles the music in both conditions? Similarly, if PolyInv truly maintains musical integrity as claimed, the polyphonic model should handle it as well as PolyOrig. But if the inversion does disrupt the music, we should see this reflected in degraded prediction scores. These metrics would validate whether the experimental manipulation works as intended. Also, how strongly are the surprisal streams correlated? There are many non-trivial modelling steps that should be reported in more detail.

    (3) Methodological inconsistencies:

    Why are the two main questions (Figures 2 and 3) answered with completely different analytical approaches? The switch from TRF to CCA with match-vs-mismatch classification seems unmotivated. I think it's very important to provide a simpler model comparison - just TRF with acoustic features plus either polyphonic or monophonic surprisal - evaluated on relevant electrodes or the full scalp. This would make the results more comparable and interpretable.

    (4) Presentation and methods:

    a) Coming from outside music/music theory, I found the paper somewhat abstract and hard to parse initially. The experimental logic becomes clearer with reflection, but you're doing yourselves a disservice with the jargon-heavy presentation. It would be useful to include example stimuli.

    b) The methods section is extremely brief - no details whatsoever are provided regarding the modelling: What specific music transformer architecture? Which implementation of this "anticipatory music transformer"? Pre-trained on what corpus - monophonic, polyphonic, Western classical only? What constituted "technical issues" for the 9 excluded participants? What were the channel rejection criteria?

  12. eLife

    Reviewer #2 (Public review):

    Summary:

    The authors sought to understand the drivers of spontaneous attentional bias and melodic expectation generation during listening to short two-part classical pieces. They measured scalp EEG data in a monophonic condition and trained a model to reconstruct the audio envelope from the EEG. They then used this model to probe which of the two voices was best reflected in the neural signal during two polyphonic conditions. In one condition, the original piece was presented, in the other, the voices were switched in an attempt to distinguish between effects of (a) the pitch range of one voice compared to the other and (b) intrinsic melodic features. They also collected a behavioural measure of attentional bias for a subset of the stimuli in a separate study. Further modelling assessed whether expectations of how the melody would unfold were formed based on an integrated percept of melody across the two voices, or based on a single voice. The authors sought to relate the findings to different theories of how musical/auditory scene analysis occurs, based on divided attention, figure-ground perception, and stream integration.

    Strengths:

    (1) A clever but simple manipulation - transposing the voices such that the higher one became the lower one - allowed an assessment of different factors that might affect the allocation of attention.

    (2) State-of-the-art analytic techniques were applied to (a) build a music attention decoder (these are more commonly encountered for speech) and (b) relate the neural data to features of the stimulus at the level of acoustics and expectation.

    (3) The effects appeared robust across the group, not driven by a handful of participants.

    Weaknesses:

    (1) A key goal of the work is to establish the relative importance for the listener's attention of a voice's (a) mean pitch in the context of the two voices (high-voice superiority) and (b) intrinsic melodic statistics/motif attractiveness. The rationale of the experimental manipulation is that switching the relative height of the lines allows these to be dissociated by imparting the same high-voice benefit to the new high-voice and the same preferred intrinsic melodic statistics to the new low voice. However, previous work suggests that the high-voice superiority effect is not all-or-nothing. Electrophysiology supported by auditory nerve modelling found it to depend on the degree of voice separation in a non-monotonic way (see https://doi.org/10.1016/j.heares.2013.07.014 at p. 68). Although the authors keep the overall pitch of the lower (and upper) line fixed across conditions, systematically different contour patterns across the voices could give rise to a sub-optimal distribution of separations in the PolyInv versus PolyOrig condition. This could weaken the high-voice superiority effect in PolyInv and explain the pattern of results. One could argue that such contour differences are examples of the "intrinsic melodic statistics" put forward as the effect working in opposition to high-voice superiority, but it is their interaction across voices that matters here.

    (2) Although melody statistics are mentioned throughout, none have been calculated. It would be helpful to see the features that presumably lead to "motif attractiveness" quantified, as well as how they differ across lines. The work of David Huron, such as at https://dl.acm.org/doi/abs/10.1145/3469013.3469016, provides examples that could be calculated with ease and compared across the two lines: "the tendency for small over large pitch movements, for large leaps to ascend, for musical phrases to fall in pitch, and for phrases to begin with an initial pitch rise". The authors also mention differences in ornamentation. Such comparisons would make it more tangible for the reader as to what differs across the original "melody" and "support" line. In particular, as the authors themselves note, lines in double-counterpoint pieces can, to a degree, operate interchangeably. Bach's inventions in particular use a lot of direct repetition (up to octave invariance), which one would expect to minimise differences in the statistics mentioned. The references purporting to relate to melodic statistics (11-14 in original numbering) seem rather to relate to high-voice superiority.

    (3) The exact nature of the transposition manipulation is obscured by a confusing Figure 1B, which shows an example in which the transposed line does not keep the same note-to-note interval structure as the original line.

    (4) The transformer model is barely described in the main text. Even readers who are familiar with the Hidden Markov Models (e.g., in IDyOM) previously used by some of the authors to model melodic surprise and entropy would benefit from a brief description in the main text at least of how transformer models are different. The Methods section goes a little further but does not mention what the training set was, nor the relative weight given to long- and short-term memory models.

    (5) The match-mismatch procedure should be explained in enough detail for readers to at least understand what value represents chance performance and why performance would be measured as an average over participants. Relatedly, there is no description at all of CCA or the match-mismatch procedure in the Methods.

    (6) Details of how the integration model was implemented will be critical to interpreting the results relating to melodic expectations. It is not clear how "a single melody combining the two streams" was modelled, given that at least some notes presumably overlapped in time.

    (7) The authors propose a weighted integration model, referring in the Discussion to dynamics and an integration rate. They do show that in the PolyOrig case, the top stream bias is highest and the monophonic model gives the best prediction, while in the PolyInv case, the top stream bias is weaker and the polyphonic model provides the best prediction. However, that doesn't seem to say anything about the temporal rate of integration, just the degree, which could be fixed over the whole stimulus. Relatedly, the terms "strong attention bias" and "weak attention bias" in Highlight 4 might give the impression of different attention modes for a given listener, or perhaps different types of listeners, but this seems to be shorthand for how attention is allocated for different types of stimuli (namely those that have or have not had their voices reversed).

    (8) Another aspect of the presentation relating to temporal dynamics is that in places (e.g., Highlight 1), the authors suggest they are tracking attention dynamically. However, as acknowledged in the Discussion, neither the behavioural nor neural measure of attentional bias are temporally resolved. The measures indicate that on average participants attend more to the higher line (less so when it formed the lower line in the original composition).

    (9) It is not clear whether the sung-back data were analysed (and if not why participants were asked to sing the melody back rather than just listen to the two components and report which they thought was the melody). It is also not stated whether the order in which the high and low voices were played back was randomised. If not, response biases or memory capacity might have affected the behavioural attention data.

  13. eLife

    Reviewer #3 (Public review):

    Summary:

    In this paper, Winchester and colleagues investigated melodic perception in natural music listening. They highlight the central role of attentional processes in identifying one particular stream in polyphonic material, and propose to compare several theoretical accounts, namely (1) divided attention, (2) figure-ground separation, and (3) stream integration. In parallel, the authors compare the relative strength of exogenous attentional effects (i.e., salience) produced by two common traits of melodies: high-pitch (compared to other voices), and attractive statistics. To ensure the generalisability of their results to real-life listening contexts, they developed a new uninstructed listening paradigm in which participants can freely attend to any part of a musical stimulus.

    Major strengths and weaknesses of the methods and results:

    (1) Winchester and colleagues capitalized on previous attention decoding techniques and proposed an uninstructed listening paradigm. This is an important innovation for the study of music perception in ecological settings, and it is used here to investigate the spontaneous attentional focus during listening. The EEG decoding results obtained are coherent with the behavioral data, suggesting that the paradigm is robust and relevant.

    (2) The authors first evaluate the relative importance of high-pitch and statistics in producing an attentional bias (Figure 2). Behavioral results show a clear pattern, in which both effects are present, with a dominance of the high-pitch one. The only weakness inherent to this protocol is that behavioral responses are measured based on a second presentation of short samples, which may induce a different attentional focus than in the first uninstructed listening.

    (3) Then, the analyses of EEG data compare the decoding results of each melody (the high or low voice, and with "richer" or "poorer" statistics), and show a similar pattern of results. However, this report leaves open the possibility of a confounding factor. In this analysis, a TRF decoding model is first trained based on the presentation of monophonic samples, and it is later used to decode the envelope of the corresponding melodies in the polyphonic scenario. The fitting scores of the training phase are not reported. If the high-pitch or richer melodies were to produce higher decoding scores during monophonic listening (due to properties of the physiological response, or to perceptual processes), a similar difference could be expected during polyphonic listening. To capture attentional biases specifically, the decoding scores in the polyphonic conditions should be compared to the scores in the monophonic conditions, and attention could be expected to increase the decoding of the attended stream or decrease the unattended one.

    (4) Then, Winchester and colleagues investigate the processing of melodic information by evaluating the encoding of melodic surprise and uncertainty (Figure 3). They compare the surprise and uncertainty estimated from a monophonic or a polyphonic model (Anticipatory Music Transformer), and analyse the data with a CCA analysis. The results show a double dissociation, where the processing of melodies with a strong attentional bias (high-pitch, rich statistics) is better approximated with a monophonic model, while a polyphonic model better classifies the other melodies. While this global result is compelling, it remains a preliminary and intriguing finding, and the manuscript does not further investigate it. As it stands, the result appears more like a starting point for further exploration than a definitive finding that can support strong theoretical claims. First, it could be complemented by a comparison of the encoding of individual melodies (e.g., AMmono high-voice vs AMmono low-voice, in PolyOrig and PolyInv conditions) to highlight a more direct correspondence with the previous results (Figure 2) and allow a more precise interpretation. Second, additional analyses or experiments would be needed to unpack this result and provide greater explanatory power. Additionally, the CCA analysis is not described in the method. The statistical testing conducted on this analysis seems to be performed across the 250 repetitions of the evaluation rather than across the 40 participants, which may bias the resulting p-values. Moreover, the choice and working principle of the Anticipatory Music Transformer are not described in the method. Overall, these results seem at first glance solid, but the missing parts of the method do not allow for full evaluation or replication of them.

    An appraisal of whether the authors achieved their aims, and whether the results support their conclusions:

    (1) Winchester and colleagues aimed at identifying the melodic stream that attracts attention during the listening of natural polyphonic music, and the underlying attentional processes. Their behavioral results confirm that high-pitched and attractive statistics increase melodic salience with a greater effect size of the former, as stated in the discussion. The TRF analyses of EEG data seem to show a similar pattern, but could also be explained by confounding factors. Next, the authors interpret the CCA results as the results of stream segregation when there is a high melodic salience, and stream integration when there are weaker attentional biases. These interpretations seem to be supported by the data, but unfortunately, no additional analyses or experiments have been conducted to further evaluate this hypothesis. The authors also acknowledge that their results do not show whether stream segregation occurs via divided attention or figure-ground separation. However, the lack of information about the music model used (Anticipatory Music Model) and the way it was set up raises some questions about its relevance and limits as a model of cognition (e.g. Is this transformer a "better" model of the listeners' expectations than the well-established IDyOM model, and why ?), and about the validity of those results.

    (2) Overall, the authors achieved most of the aims presented in the introduction, although they couldn't give a more precise account of the attentional processes at stake. The interpretations are sound and not overstated, with the exception of potential confounding factors that could compromise the conclusions on the neural tracking of salient melodies (EEG results, Figure 2).

    Impact of the work on the field, and the utility of the methods and data to the community:

    The new uninstructed listening paradigm introduced in this paper will likely have an important impact on psychologists and neuroscientists working on music perception and auditory attention, enabling them to conduct experiments in more ecological settings. While the attentional biases towards melodies with high-pitch and attractive statistics are already known, showing their relative effect is an important step in building precise models of auditory attention, and allows future paradigms to explore more fine-grained effects. Finally, the stream segregation and integration shown with this paradigm could be important for researchers working on music perception. Future work may be necessary to identify the models (Markov chains, deep learning) and setup (data analysis, stimuli, control variables) that do or do not replicate these results.

  14. Mentioned by @biorxiv-neursci.bsky.social on Bluesky

    Where is the melody? Spontaneous attention orchestrates melody formation during polyphonic music listening https://www.biorxiv.org/content/10.1101/2025.08.26.672294v1

  15. Mentioned by @biorxivpreprint.bsky.social on Bluesky

    Where is the melody? Spontaneous attention orchestrates melody formation during polyphonic music listening https://www.biorxiv.org/content/10.1101/2025.08.26.672294v1

  16. Version published to 10.1101/2025.08.26.672294v1 on bioRxiv