What grounds singing voice preferences?
Mon-HS1-Talk II-01
Presented by: Camila Bruder
In the visual domain, the relative contribution of object features and viewers' idiosyncrasies to aesthetic judgments is often discussed. In a recent study, we focused on singing performances and examined the contribution of attributes of the voice signal to listeners' liking of pop performances. Concretely, we asked participants to report how much they liked melodies (a cappella excerpts of “Don’t worry, be happy” and “Over the rainbow”), performed by highly trained female singers. The material was also described in terms of various perceptual features (e.g., vibrato, articulation, tempo etc.) by the same participants and acoustically analysed (e.g., fundamental frequency, jitter, shimmer, etc.). Results indicate that liking can be predicted by perceptual features of the voices (which account for 44% of the variance in liking ratings), but not by their acoustic features. To generalize our findings, we recently constructed a dataset of human vocalizations, consisting of recordings of 22 singers performing six melodies in three different singing styles (as a lullaby, as a pop song, as an opera aria). The dataset has been validated in a forced-choice lab experiment where lay listeners (N = 25 for each stimulus) could discriminate if stimuli sounded as a lullaby, a pop song or an opera aria with mean accuracy higher than 75% for all three singing styles. By using this rich new stimulus dataset in listening experiments, we are currently exploring listeners’ singing voice preferences in a wider context and ultimately drawing a parallel between the visual and auditory domains.
Keywords: aesthetics, liking, acoustics, singing voice, preferences