Open Access

Singing is ubiquitous and highly significant to human experience. Yet, despite the relevance and popularity of singing, scientific inquiry into singing voice preferences remains relatively scarce. This dissertation provides a rigorous empirical investigation of singing voice preferences. These were studied with an interactionist approach, considering aspects of the performances themselves, the listener who makes the evaluations, and the interaction between them. The first study aimed to quantify how much singing voice preferences could be linked to attributes of the singing performances. Modeling of participants’ liking ratings showed that they could be partially predicted by participants’ perceptual ratings of the stimuli, but not by computationally extracted representations of the acoustic signal. Importantly, there were large individual differences both in participants’ preferences and in how participants perceived the voices, as shown by the low interrater agreement in both liking and perceptual ratings. To generalize findings to other styles of singing and integrate them into the relatively abundant literature on speaking voice attractiveness, a new stimulus set of contrasting vocalizations was recorded. In the CoVox dataset, the same musical/text material was performed by 22 highly trained, versatile classical singers in three singing styles (operatic, pop, and lullaby singing) and two speaking styles (infant- and adult-directed speech). The CoVox dataset was carefully validated and described in a second study, and is openly available to the research community. In a third study, the singing performances of the CoVox dataset were examined with a focus on the relationship between singers’ versatility – determined by how well the intended style of singing performances could be recognized – and singers’ formal musical training. Singers’ versatility could not be linked to their musical training, a finding relevant to voice pedagogues concerned with teaching healthy “cross-over abilities” to their singing students. Finally, aesthetic preferences for contrasting vocalizations were examined using a subset of the CoVox dataset. In a registered report, a parallel with the visual domain inspired predictions concerning the amount of shared taste elicited by the five vocalization styles: it was expected that lullabies, as a more “natural” or universal category of singing, would lead to higher shared taste than operatic singing, a more “artificial” category of singing, with pop singing in an intermediate position. Contrary to predictions, lullaby and operatic singing led to equivalent shared taste, higher than what was found for pop. This registered report also tested the “singing and speaking as backup signals” hypothesis, according to which the attractiveness of voices is based on redundant acoustic cues of individual fitness. Results did not support the hypothesis of highly consistent average preferences for the same singers across all vocalization styles, suggesting that, if sexual selection plays a role in singing voice preferences, that role is limited. This dissertation employed an integrative approach to characterize voice preferences. Findings demonstrated that, while listeners' preferences can be consistent over time, interrater agreement is typically very low, which converges with the finding that (objective) acoustic characteristics poorly predict singing voice preferences. The highly idiosyncratic nature of voice preferences makes them a challenging topic to study. Nevertheless, the studies reported here establish important conceptual and methodological foundations for future research.