1 of 1

EEG Spectral Analysis of Valence Neural Encoding in Narrative Listening

Deepa Tilwani1-4, Xuan Yang4,5, Svetlana V. Shinkareva4,5, and Christian O'Reilly1-4

​

1 Laboratory for Integrative Neuroscience Analysis (LINA), Department of Computer Science, University of South Carolina.

2 AI Institute, University of South Carolina.

3 Carolina Autism and Neurodevelopment Research Center, University of South Carolina.

4 Institute for Mind and Brain, University of South Carolina

5 Department of Psychology, University of South Carolina.

Introduction

  • Emotions are fundamental to human experience, influencing perception, decision-making, and social interactions
  • Psychiatric disorders like depression and anxiety involve dysregulated affective processing
  • EEG offers superior temporal resolution and direct measurement of neural activity
  • Hedonic valence, representing the pleasantness or unpleasantness dimension of emotional experience. EEG allows us to track its rapid neural dynamics with millisecond precision, revealing how affective states unfold across time and frequency.

Objective

To characterize the spectral fingerprints of hedonic valence during continuous narrative listening using high temporal resolution of EEG.

Methods

Results

  • Top Regions (Fig 1A): Insula, superior temporal, and ventrolateral prefrontal cortices exhibit the strongest PSD–valence correlation, especially in the β and high‑γ bands.
  • Band Summary (Fig 1B): Correlation strength peaks in the β band, followed by γ, α, then θ.
  • Hemispheric Bias (Fig 1C): Right‑hemisphere ROIs show a modest advantage over left.
  • Example Spectra (Fig 1D): Top three ROIs for left and right hemisphere.
  • ROI Categories (Fig 2A): The violin plot showing the correlation strength distribution across different brain region categories. It shows ROI brain regions have 3.7% stronger PSD-valence correlations than non-ROI areas.
  • Frequency Comparison (Fig 2B): Different brain regions correlate with valence for the (in decreasing order) β, γ, α, and θ frequency bands.
  • ROI Distribution (Fig 2C): The count of affective ROI with significant correlations are well balanced across hemispheres, with many regions per category.
  • Top by Category (Fig 2D): Within each ROI category, exemplar regions (e.g., anterior cingulate, (s32)) show significant correlations.
  • PSD–valence relationships were too variable to draw clear conclusions for all participants. A surrogate‑based threshold isolated 11 participants with consistently strong coupling, enhancing signal reliability but introducing selection limits.
  • Among 11 participants, insula and superior temporal cortex showed modest β‑band elevations.
  • A modest right‐hemisphere bias aligns with known prosodic lateralization but remains descriptive; rigorous statistical testing in the full sample is required before claiming true lateralization.
  • Lack of strong coupling for all participants will require future investigation.

Discussion

β‑band EEG power in key affective networks may track moment-to-moment valence in a subset of participants, suggesting the pipeline’s potential while underscoring the need for broader validation and future analyses of age‑ and sex‑related influences on valence representation.

​

Conclusion

Acknowledgements

This work was supported by the National Science Foundation (Award #2419634). We thank the ABC@USC cohort participants and ABC@USC research staff for their invaluable contributions.

Auditory Listening Task (8 min total)

6-min English Narrative

50s Hindi

50s Thai

A linguist segmented the 6-min English narratives

​

Segments

High

Positive

Low

Negative

​

Statistical Validation of Selected Participants: To assess reliability, correlations between power spectrum density (PSD) and valence were bootstrapped (100 iterations) by resampling data segments with replacement. Surrogate analyses were also performed by randomly shuffling valence labels to generate a null distribution. Comparing real and surrogate results allowed identification of consistent and statistically significant brain–emotion associations.

​

Affective Ratings of Narrative

EEG Pipeline

Participants listen attentively to the narrative, MRI and EEG recordings collected throughout the listening task.

Data Collection

L_32: anterior cingulate

L_STSda: superior temporal sulcus

L_AAIC_a: anterior insula

R_V2: second visual area

R_STGa: anterior superior temporal gyrus

R_52: primary auditory cortex