Speaking Style Based Apparent Personality Recognition

摘要

In this study, we investigate the problem of apparent personality recognition using person's voice, or more precisely, the way he or she speaks. Based on the style transfer idea in deep neural net image processing, we developed a system capable of speaking style extraction from recorded speech utterances, which then uses this information to estimate the so called Big-Five personality traits. The latent speaking style space is represented by the Gram matrix of convoluted acoustic features. We used a database with labels of personality traits perceived by other people (first impression). The experimental results showed that the proposed system achieves state of the art results for the task of audio based apparent personality recognition.