Special Talk - "Investigating the Memorability of Voices "

Abstract: Memorability, or the likelihood of a stimulus to be remembered, is an intrinsic property of stimuli that is highly consistent across viewers. In other words, people tend to remember and forget the same faces (Bainbridge et al., 2013), scenes (Isola et al., 2014), objects (Kramer et al., 2023), and more. However, memorability research until now has been entirely limited to the visual domain. We provide the first exploration of auditory memorability, investigating whether this consistency in what individuals remember and forget extends to speakers’ voices, and if so, what makes a voice memorable. Over 2500 online participants performed a continuous recognition task in which they heard a sequence of the same sentence read by different speakers from a largescale speech database (Garofolo, 1993), and indicated whenever they heard a repeated voice. We found that participants were indeed significantly consistent in their memory performance for voice clips. Interestingly, memorability scores (d’) for sentences spoken by the same speakers were significantly correlated, suggesting consistent memorability at the level of speakers beyond specific utterances. Next, we ran regression models incorporating both low-level properties (e.g., pitch, harmonic amplitude) and high-level properties (e.g., dialect, perceived confidence) of the stimuli to predict their memorability. The final models, which contained primarily low-level predictors, were significantly predictive of d’. Critically, these models successfully cross-validated to clips of another sentence spoken by the same speakers, providing further evidence that the memorability of speakers is intrinsic and predictable. Overall, our findings can be used to control for the memorability of stimuli in auditory experiments, and more generally shed light on the processes of speech perception and memory.