How Sweet the Sound? Modeling Acoustic Correlates of Timbral Sweetness

Sweetness has been a term long associated with music, commonly used to describe melodies, harmonies, and timbres. The current study aims to identify acoustic correlates and explore musicians’ conceptions of timbral sweetness. Six professional musicians representing eight different instruments recorded single, sustained notes in two conditions: in one, they were instructed to play “with as much sweetness in your tone as possible,” and in the other, “with as little sweetness in your tone as possible.” Following the recording session, we interviewed participants about their experience and their understanding of timbral sweetness. We report here the results of audio descriptor analysis of the recordings as well as content analysis of the interviews. We used random forest models to classify the playing condition (sweet v. not sweet) based on audio descriptor input. Our final model includes 13 features, with an accuracy of 70% (compared to 50% chance). Analyses from both a preliminary study and the main study offer converging support for the roles of features suggested by previous research, including inharmonicity, noisiness, spectral centroid, attack slope, and fundamental frequency (F0). Our results suggest that spectral, temporal, and spectrotemporal features all play crucial roles in interpreting sounds as “sweet.” Qualitative analysis of the interviews revealed five primary themes: (1) Focus and Core, (2) Dark, Warm, Rich, and Round, (3) Liveliness and Youth, (4) Less Pressure and Effort, and (5) Vibrato and Legato. Participants did not spontaneously describe timbral sweetness in gustatory or olfactory terms, suggesting that value, emotion, and/or qualitative properties may drive cognitive interpretations of timbral sweetness over crossmodal correspondence.