The vocal uncanny valley: Deviation from typical organic voices best explains uncanniness.

Abstract

Abstract The uncanny valley describes the negative evaluation of near humanlike artificial entities. Previous research with synthetic and real voices failed to find an uncanny valley of voices. This may have been due to the selection of stimuli. In Experiment 1 (n = 50), synthetic, normal, and deviating voices (distorted and pathological) were rated on uncanniness and human likeness and categorized as human or non-human. Results showed a non-monotonic function when the uncanniness was plotted against human likeness indicative of an uncanny valley. However, the shape could be divided into two monotonic functions based on voice type (synthetic vs deviating). Categorization ambiguity could not predict voice uncanniness but moderated the effect of realism on uncanniness. Experiment 2 (n = 35) found that perceived organicness of voices significantly moderated the effect of realism on uncanniness, while attribution of mind or animacy did not. Results indicate a vocal uncanny valley re-imagined as monotonic functions of two types of deviations from typical human voices. While voices can fall into an uncanny valley, synthetic voices successfully escape it. Finally, the results support the account that uncanniness is caused by deviations from familiar categories, rather than categorical ambiguity or the misattribution of mind or animacy.