Speaker Accommodations towards VUI Voices on the dimensions of Voice Onset Time and Pitch Range

Abstract

Previous research has found evidence of speech accommodation between human interlocutors [Lindblom, B. (1990). Speech Production and Speech Modeling] particularly in the dimensions of voice onset time (VOT) [Asano, Y., & Gubian, M. (2018). Speech Communication] and pitch range [Oviatt, S., Maceachern, M., & Levow, G.-A. (1998). Speech Communication]. With the growing presence and integration of voice user interfaces (VUI), inanimate devices which use human voices, the question arises of whether accommodation is present when the interlocutor is a VUI, and whether perceived animacy further influences this. Thirteen prompts were read by twenty-seven participants over Zoom who then heard synthetic speech responses from two voices believed to be active VUIs. These voices were produced using Amazon AWS Polly and differed in perceived animacy, a “human-like voice” and a “robotic voice” each manipulated to have shorter and longer VOT and different pitch range respectively. Pitch and VOT of the participants’ directives were analysed. Current findings suggest that participants do not accommodate VOTs to either voice, and that participants’ pitch patterns diverge from the VUIs pitch range. These preliminary results suggest that phonetic characteristics are employed differently in communication towards VUIs.