Publikation
Parametric Affect: An Empirical Study on Pitch and Tempo in Generative Music
Özlem Salman; Ko Watanabe; Javier Melo; Andreas Dengel; Shoya Ishimaru (Hrsg.)
International Joint Conference on Pervasive and Ubiquitous Computing (UbiComp-2026), October 11-15, Shanghai, China, ISBN 979-8-4007-2533-3/2026/10, Association for Computing Machinery, 2026.
Zusammenfassung
Generative AI is increasingly embedded in everyday audio experiences, from personalized playlists to text-to-music tools that synthesize compositions on demand. Yet little is known about how listeners emotionally perceive these AI-generated pieces, or which acoustic features lead listeners to judge a piece as human- or machine-made. We report an exploratory listener study (𝑁 =21) using 24 five-second clips generated with MusicGen, twelve prompted as “happy” and twelve as “sad”. Participants rated each clip on emo- tional valence, the perceived influence of pitch and tempo, and whether it sounded human-made or AI-generated. Participants recovered the intended emotion in approximately 79% of trials, and tempo ratings were systematically associated with: high-tempo clips were more often labeled happy and more often attributed to a human composer. Pitch effects were statistically significant but small. We discuss implications for designing emotionally and socially appropriate generative audio in ubiquitous computing settings.