A Group Listening Experiment on Authorship Discrimination in AI-Generated and Human-Produced Music Across Three Instrumental Genres
Year: 2026
Type: conference-paper
Abstract: This study examines whether authorship discrimination between AI-generated and human-produced music remains statistically meaningful when listening occurs as a group classroom experiment over shared loudspeakers rather than in controlled, individual headphone sessions. Building on Longardner (2026), undergraduate and graduate-level music students at the Sichuan Conservatory of Music (N = 107) completed a forced-choice authorship task on 30 instrumental excerpts spanning Lo-Fi, Hard Rock, and Swing Jazz. The stimulus set included 18 AI-generated tracks produced with Suno, Udio, and Mureka, as well as 12 human-produced tracks drawn from studio-quality releases. Across 3,210 responses, overall identification accuracy was 48.16%, statistically indistinguishable from chance, indicating that participants could not reliably differentiate AI-generated from human-produced recordings under group listening conditions. These findings suggest that listening context meaningfully constrains perceptual discriminability and that classroom-based AI literacy activities should not assume consistent authorship-identification performance in the absence of controlled playback conditions.
Author: Dr. Longardner, Joseph
Institution: Sichuan Conservatory of Music
Language: English
Topic: Music and Audio Processing
Subfield: Signal Processing
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7207530972