Speech Enhancement Patterns in Human-Robot Interaction: A Cross-Linguistic Perspective

This paper presents the results of the human-robot interaction (HRI) study with German native speakers addressing the robot in their L1 and in L2 English. The aim of the experiment is to test the strategies of providing clarifications when talking to the voice assistant in a task involving teaching complex vocabulary. The analyses is based on spectral (F1, F2, and mean F0) and temporal (vowel length) features excerpted from the target words. With reference to a theoretical framework of hyperarticulation and hypoarticulation, these acoustic measures were compared across the iterations of the target words (first vs. second iteration). Results showed that participants, when asked for clarification by an inanimate interlocutor, do not hyperarticulate, but try to preserve the surface representation of target words across the iterations. These findings suggest that acoustic characteristics of clarifications directed to voice assistants differ from the ones directed to human interlocutors.