acceptodds
Under review as a conference paper at ICLR 2027

NATURALISTIC MUSIC STABILIZES ALPHA-BAND PHASE DYNAMICS IN HUMAN ELECTROENCEPHALOGRAM

Abstract

The temporal evolution of neuronal phase and its predictability in naturalistic settings, such as music listening, remains underexplored in the alpha-band (8-13 Hz) of electroencephalogram (EEG) studies. In this study, we propose a novel framework for quantifying short-horizon phase predictability and phase stability by modeling phase prediction errors using a linear approximation that captures the diffusion-like behavior of alpha-band oscillations. The framework was evaluated on two music-EEG datasets. The first dataset consists of EEG data from 13 healthy participants listening to live Indian classical music (ICM), comprising two ragas and relaxation conditions. The second dataset, the naturalistic music EEG dataset - tempo (NMED-T), consisted of EEG recordings from 20 participants listening to 10 songs. For both datasets, individual alpha frequencies (IAF) were computed to obtain subject-specific band-pass signals. Instantaneous phase was extracted from the analytic signal using the Hilbert transform and predicted over short horizons (Δ = 50, 100, 150, 200 ms) using instantaneous frequency estimates. Mean absolute phase prediction errors were modeled linearly to estimate the diffusion slope, providing a measure of phase stability. The results from the ICM dataset showed consistently higher phase prediction errors during the relaxation condition than during both music conditions. The diffusion slopes were lower during music, suggesting slower accumulation of phase prediction error and greater short-horizon phase stability. Paired t-tests revealed statistically significant differences between music and relaxation for both phase prediction errors and diffusion slopes. Comparative results were observed on the NMED-T dataset. These findings demonstrate that naturalistic music listening is associated with more stable alpha-band phase dynamics, providing a quantitative framework for assessing neural oscillatory stability in continuous EEG signals.

Primary area:
applications to neuroscience & cognitive science
Keywords:
Electroencephalogram, Alpha oscillations, Phase predictability, Diffusion slope, Hilbert transform, Naturalistic music, Individual alpha frequency

Reviews

Among:
Shape:
1234this paper, Bottom 1%
Figure 1. Mean review rating of the reviewed papers at ICLR 2027. This paper averages 1.00 over 3 reviews, above 0% of the others.
Reviewer_CSL8OpenReview ↗
Rating
1
Confidence
2
Informativeness
2
Presentation
2
Soundness
2
Read review
Paper summary

This study investigates the short-term evolution of the phase of alpha-band oscillations during continuous naturalistic conditions (music listening) on two datasets of EEG signals. The first dataset comprises EEG signals from13 healthy participants listening to live Indian classical music, while the second dataset consists of EEG signals measured from 20 participants listening to 10 songs. Based on the analysis done by the authors, the main conclusion is that the naturalistic music listening is associated with more stable alpha-band phase dynamics.

Critical strengths

S1. The topic addressed in this study is of interest for the researchers working in neuroscience.

Critical weaknesses

W1. The paper does not really propose a novel method, but rather a combination of existing techniques.

W2. The paper is only focused on a single niche application.

W3. The empirical validation is not rigorous (see the questions below).

W4. The code used in experiments does not seem to be publicly available.

Questions for authors

Q1. The short-horizon phase prediction framework relies on a first-order linear extrapolation derived from the instantaneous frequency. Why the standard methods from the time series literature are not used as baselines? What are the guarantees that the estimated diffusion slope is not an artifact of the application of a naive linear predictor?

Q2. The first dataset (ICM) consists of EEG signals recorded from 13 participants. Is a sample size of N=13 statistically sufficient for a benchmark? The second dataset (NMED-T) contains EEG signals from 20 participants listening to 10 songs. What was the reason for randomly selecting only 2 out of 10 songs from NMED-T?

Reviewer_yCLMOpenReview ↗
Rating
1
Confidence
3
Informativeness
2
Presentation
3
Soundness
2
Read review
Paper summary

This paper investigates alpha-band phase predictability during naturalistic music listening by extrapolating Hilbert phase using instantaneous frequency estimates.

Critical strengths

The proposed method is simple and straightforward to understand. Evaluating multiple prediction horizons and applying the pipeline to a second naturalistic EEG dataset provide useful preliminary evidence regarding alpha-phase dynamics during music listening.

Critical weaknesses

The diffusion interpretation is insufficiently supported, as the analysis fits a linear model to mean absolute phase errors at only four prediction horizons rather than examining the theoretically expected growth of phase-increment variance. Furthermore, differences between conditions may reflect variations in signal quality that affect Hilbert phase and instantaneous-frequency estimation, producing larger prediction errors without necessarily indicating changes in neural phase stability. Importantly, zero-phase filtering and the standard Hilbert transform incorporate future samples, making the analysis noncausal. Consequently, the proposed procedure primarily characterizes offline local phase smoothness rather than real-time phase forecasting available at time t. The claims and terminology should be moderated accordingly.

Questions for authors
  • What evidence supports the diffusion interpretation beyond linear fitting of mean absolute errors at four prediction horizons?

  • Can the authors report goodness-of-fit statistics and compare alternative nonlinear error-growth models?

  • Can the analysis be repeated using causal filtering and online phase estimation to assess genuine forecasting performance?

Reviewer_5dX3OpenReview ↗
Rating
1
Confidence
3
Informativeness
1
Presentation
3
Soundness
3
Read review
Paper summary

The paper investigates the error of an instantaneous phase estimate based on instantaneous frequency through linearization in the alpha-band of a spatially averaged EEG signal during listening to music or relaxation. The instantaneous estimate is based on the Hilbert transform, unwrapped phase of the band-pass filtered signal where the center frequency is based on an individualized estimate of alpha. The distribution of mean absolute angular error is a function of lag is then fit with a linear model to understand the slope. This is a related to the diffusion of phase for a noisy oscillator (where the angular error is a proxy for the phase variance).

Critical strengths

The paper is clear and there is no substantial errors.

Critical weaknesses

The overall contribution of the paper is limited. The number of datasets is small and while the statistics are signifiant, the paper's significance in terms of insight is lacking. The fact that broadband noise (white for example) will have more impact on instantaneous alpha when there is lower alpha power (SNR is lower) makes the results in Table 2 quite obvious from Figure 2.

Questions for authors

More of a comment rather than a question: Figure 1 is quite large and doesn't add information beyond what is already described in the text.

Isn't it the case that testing the 6 different hypotheses (pairs at Δ=100\Delta=100 and then slope) are dependent?

Minor point: Line 44 (and elsewhere) use parenthetical references instead of inline. It could be better explained that the local linear estimation 7 used instead of variance because there is not multiple realizations.

Title capitalization in the venues (journal or conference) are not applied consistently in the reference list.

From OpenReview, CC BY 4.0. Authors as signed there; shown unedited.

Related papers

Loading the map…

Discussion (0)

Sign in to comment.