Dissertations, Theses, and Capstone Projects

Date of Degree

6-2026

Document Type

Master's Thesis

Degree Name

Master of Science

Program

Cognitive Neuroscience

Advisor

Edward A. Vessel

Subject Categories

Cognition and Perception | Cognitive Psychology | Cognitive Science | Other Music

Keywords

temporal integration, music cognition, aesthetic judgment, continuous ratings, transformer attention, audio spectrogram transformer

Abstract

Music unfolds over time, requiring listeners to integrate information moment by moment to make aesthetic judgments. Although we know that music is likely processed on multiple timescales, it is not known which musical features contribute to short versus long timeframes. This study examined whether a computational measure of temporal integration could help explain aesthetic responses to music. Participants (N = 28) listened to 32 sixty-second clips of classical and electronic music, and continuously rated their enjoyment with a dial. After listening, they then gave an overall liking judgment for each clip. To quantify temporal integration, we derived an integration measure known as Mean Attention Distance (MAD) from the pretrained transformer model Audio Spectrogram Transformer (AST). This metric, which was turned into an early and late layer MAD score for each piece, was used as an estimate of how the model integrated information over time for each excerpt.

Analyses compared early and late layer MAD scores across genres, and tested whether MAD predicted aesthetic judgments and listener agreement. Early MAD was similar across both genres, suggesting that the timeframes of local integration were similar across genres. By contrast, Late MAD was numerically higher for electronic clips than classical clips, although this v genre difference did not reach statistical significance. Across regression models, late MAD emerged as more informative, while early MAD was less related to behavioral outcomes. These findings suggest that more extended temporal structure, rather than local integration alone, may be relevant to how listeners evaluate music, but did not reliably explain listener agreement across responses. More broadly, this study supports the use of attention measures from transformer architectures as an approach for studying temporal structure in aesthetic experience.

Download

Share

COinS