Judgements of a customer service agent's personality based on acoustic properties of their speech

Abstract

People frequently interact with customer service agents, whether they are a real person or an AI, and quickly form personality impressions from it. The present study aimed to investigate how acoustic characteristics of speech of human agents affect their personality impressions. To this end, spontaneous speech was elicited from 30 native Korean speakers while they played a role of an agent assisting a customer on the phone. We then conducted a personality rating experiment in which 30 native speakers of Korean rated the perceived degree of extraversion, agreeableness, conscientiousness, neuroticism, and openness (the “Big Five” personality traits) on the recordings on a 5-point scale. Various acoustic features were extracted from the stimuli, which were converted into two principal components using a Principal Component Analysis. The results of correlation analyses demonstrated that some of the acoustic variables such as measures of voice quality (H1-H2, Harmonic-to-Noise Ratio), the mean f0, and articulation rate were significantly correlated with the personality ratings, with the acoustic correlates different for male and female speakers. This helps discover characteristics of a more likable voice and speaking style for a customer service agent and has important implications for generating more likable synthetic speech for similar services.