Automatic Personality Recognition from reading text speech

Abstract

Study on personality trait has been a very active research field in psychology for long time. In recent years automatic identification of the speaker personality has attracted the attention of many researchers and Five-Factor Model of Personality has become the predominant model of general personality structure. The results presented in this paper come from a study of how individual's reading text speech manifest personality. The evaluation of proposed methods are carried out on reading text speech of 140 subjects and NEO-FFI questionnaire's scores. For feature extraction, we used popular OpenSMILE toolkit and available ComParE 2013 audio feature set. For classification, we used Linear Kernel SVM classifier together with five filter feature selection techniques and Principal Component Analysis. Each SVM trained individually for each trait with repeated cross validation and five feature sets. Best achieved UAR has range from 74% to 80% depends on different traits.