Poster Presentation

Poster B in Poster Session B: Tuesday, August 4, 2:00 – 3:45 pm, Kimmel Center, Shorin & Rosenthal Rooms

Using Transformer Attention to Assess How Temporal Integration in Music Relates to Aesthetic Pleasure

Edward A Vessel¹, Sophia Senderak², August Miller³; ¹CUNY City College of NY, ²CUNY Graduate Center, ³The City College of New York

Presenter: Edward A Vessel

Music unfolds over time, requiring listeners to integrate information across successive moments. How does this integration relate to a listener's aesthetic enjoyment? Previous studies have shown that listeners track such long-range temporal dependencies and that they influence enjoyment; however, these measurements have been restricted to simple melodies for which probabilistic models can be calculated. Here, we study how long-range musical structure relates to aesthetic enjoyment in actual musical recordings. Thirty participants listened to 60 sec clips of music from two genres (electronic, classical) while using a dial to report moment-by-moment enjoyment, followed by an overall aesthetic judgment of the full clip. The presence of long-range temporal structure in each clip was quantified using a computational measure of context dependence. This measure,mean attention distance (MAD), was derived from attention matrices of a transformer architecture adapted for use with music. A major advantage is that MAD can be computed directly from audio files. Initial analysis of N=19 participants reveals differences in MAD scores across musical clips and a potential relationship between MAD scores and peak enjoyment.

Topic Area: Auditory, Speech & Language Processing