Perception of Speaker Personality Traits Using Speech Signals

As conversational agents continue to replace humans in consumer contexts, voice interfaces must reflect the complexity of real-world human interaction to foster long-term customer relationships. Perceiving the personality traits of others based on the way they look or sound is a key aspect of how humans unconsciously adapt their communication with others. In an effort to model this complex human process for eventual application to conversational agents, this paper presents the results of (1) building SVM and HMM classifiers for perceived personality prediction using speech signals using a data corpus of 640 speech signals based on 11 Big Five personality assessments, (2) determining correlations between feature and speaker subgroups, and (3) assessing the SVM classifier performance on new speech signals collected and assessed through a user study. This work is a small step towards the greater goal of designing more emotionally intelligent conversational interfaces.