Toward CI-Accessible Music AI: Emotional Perception and Representation of AI-Generated Music for Cochlear Implant Users

Taein Song, Hyunjae Kim, Kyung Myun Lee

Submission Type: 2-page Extended Abstract (Non-archival) / 2페이지 Extended Abstract (프로시딩 미수록)

Keywords: Cochlear Implants, Musical Emotion Perception, AI-Generated Music, Music Foundation Models, Accessibility, Human–Model Alignment

Abstract:

The perception of diverse emotions in music, driven by multiple acoustic cues, plays a central role in human life. However, for Deaf and Hard-of-Hearing (DHH) populations, including cochlear implant (CI) users, degraded acoustic cues can limit access to musical emotion. Motivated by recent advances in generative AI, we investigate its potential to improve music accessibility for these populations. We conducted a behavioral experiment in which CI and normal-hearing (NH) groups rated perceived emotions of human-composed and AI-generated music. Moreover, under the same stimuli (human vs. AI) and listening conditions (NH vs. CI), we analyzed foundation model representations to examine how musical emotions are encoded and align with human responses.
Behaviorally, emotion categories were less separable for AI-generated music than for human-composed music. A similar reduction was observed in the CI group compared to NH group, resulting in the lowest separability in the CI–AI condition. Foundation models showed reduced prediction accuracy for ratings in the CI group, and exhibited similar patterns of emotion separability as observed in the behavioral experiment. These results suggest that current AI-generated music fails to fully capture emotional structure, particularly under degraded listening conditions, highlighting the need for improved accessibility for diverse auditory populations.

Email Sharing: We authorize sharing author emails with Program Chairs. / 저자 이메일 공유에 동의합니다.

Data Release: We understand and agree to the OpenReview metadata visibility policy. / OpenReview 메타데이터 공개 정책을 이해하고 동의합니다.

Submission Number: 23