An Experimental Approach to the Perception of Empathy in Speech

With advances in the techniques and naturalness of speech synthesis, and the increasing commercial contexts in which it is used, such as in personal assistants, the need for natural affective synthesis has grown, allowing for readier incorporation of socially-intelligent agents into society. Although affective synthesis is by no means a new field, there is a need now more than ever to understand the acoustic correlates of emotions in natural speech to optimise this synthesis. Plenty of research has been conducted relating to so-called ‘primary emotions’ such as anger or sadness, but less work has been done on more subjective and socially-conditioned ‘secondary emotions’ like empathy. As a means of engaging with social robots, empathy is particularly salient due to its nature of demonstrating emotional understanding and engagement, and has clear use in medical technology (James et al., 2018) and social robots (Asada, 2015). This original research seeks to shed more light on the nature of empathy in speech, using a parametric approach to synthesis. Although an older technique compared to machine learning synthesis, it is found that this approach allows for a greater degree of control over acoustic correlates, and allows for a more precise image of empathy to emerge. This research splits empathy into production and perception; the first experiment looks at the difference between non-empathetic and empathetic contexts in a dialogue context with 10 participants (5 male and 5 female, in pairs), which allows for the identification of empathetic correlates in natural speech production. Following this, an experiment using resynthesised versions of the non-empathetic productions explores which combinations and amounts of the correlates observed in the first experiment (including pitch, duration, and voice quality) must be used in order to elicit empathy. In doing so, it becomes clear that empathy is not a unified concept in speech, instead behaving differently within and between production and perception.

This individual article from the Proceedings is published here.