Abstract
Automatic personality recognition using the Big Five dimensions (OCEAN: extraversion, agreeableness, conscientiousness, neuroticism and openness) is capturing the attention of researchers. Personality recognition is expected to have encouraging future in Human- computer and Robot Interaction applications. Human speech conveys rich information that can be derived to recognize speaker traits. However, our focus is on the rich content of non-verbal features in human speech. We focus on how humans talk, not what they talk about. The focus in this paper is to experiment with four different machine learning techniques, and their performance in recognizing personality traits, we report our results in this regard. We use the Speaker Personality Corpus provided by the Interspeech 2012 challenge. First, we recognize three issues affecting the system's low performance: dimensionality, judges' agreement, and imbalanced data. Next, we address each issue and provide a solution to improve the system's performance. Finally, we compare our results with the baseline showing better classification results.