Estimating Personality Impression from Speed Record Using Hidden Markov Models

摘要

When people listen to other's speech for the first time, they always attribute personality traits to the speaker subconsciously. We consider that if robots can predict personality traits of users from their speech, the communication in human-machine interaction will improve significantly. This paper proposes an approach for the automatic estimation of the traits, in which the listeners attribute to unanimous speakers. And the discrimination experiments based on hidden Markov model and canonical discrimination analysis show that, it is possible to predict with high accuracy (more than 75%), whether a speaker is perceived to be in the higher or lower part of the "Extraversion", "Openness", and "conscientiousness" by using nonverbal information.