Personality in Speech

This chapter gives an introduction into the understanding and assessment of personality. While many perspectives and models have been proposed in the literature in order to capture or estimate personality, the chosen approach underlines the empirical character of the phenomenon at hand elaborating on on the Trait Theory of personality, its origin, application and implications as well as the chosen NEO-FFI questionnaire for personality assessment throughout this work.

As the previous chapter outlined approaches to personality assessment in psychology, this chapter summarizes works and insights of researchers from the speech community. Many of these researchers are linguists or computer scientists, hence the aim of approaching an individual’s personality translates into the aim of modeling or experimenting with personality. Essentially, the assessment of perceivable manifestations of personality is the basis for any experimentation. When analyzing personality in terms of speech, the scope of interest is narrowed down from overall personality, i.e., maybe being able to judge about personality from previous knowledge about actions or incidents, towards focusing on perceivable characteristics, in this respect it means perceivable at the very point in time the conversation or the experiment occurs as well as comprehensible to any person including persons having no prior knowledge about the speaker. Resulting limitations and cleavages of this respective will be addressed in the present chapter.

This chapter explains the three databases recorded for this work as well as the labeling procedure. The first set of recordings comprises speech that is produced by a single professional actor who was given a fixed text passage to portray out of different personality perspectives. For a second series of records the actor was invited several times with recording sessions spread over several weeks. This time, the actor was presented images like photos, drawings or artwork. The images were selected to trigger emotions and feelings. Eventually, the actor was asked to speak freely about any associations when looking at the images while acting out of different personality perspectives. Because of database design this set provides the opportunity to analyze time- and text-dependency in personality expression at the same time. For a third series of recordings 64 native speakers were invited. The speakers were engaged in different conversational scenarios, no personality perspectives were induced. In addition, this set comprises two technical recording conditions, as half of the speakers were recorded with stand alone microphones while the other half were recorded with a headset. In order to generate personality labels listening tests were conducted for a range of selected stimuli using full NEO-FFI questionnaires.

This chapter provides exploratory insights into the personality-related expressions and interdependencies in the datasets. At the same time, the question of whether or not the chosen assessment scheme can be applied is analyzed. Eventually, the discussion of the overall high consistencies, the observation of normal distributions in the ratings, the comparable correlation patterns in between the traits, the very congruent latent factor structure as well as the significant differences in between the target groups show that the induced personality expressions are perceived as intended by human listeners.

This chapter describes how personality cues can be estimated from speech using an automated system. In order to enable a machine to come to a decision or estimation about any speakers’ personality a comprehensive processing chain needs to be developed. First, the recordings need to be preprocessed and segmented into meaningful chunks. Next, promising acoustic or prosodic cues need to be extracted from the signal. Because the realization of a feature candidate to be ‘promising’ can be expected to vary between the traits, this work incorporates a generic feature selection scheme in order to identify promising features from a completely data-driven perspective, namely the IGR ranking algorithm. Using promising features discriminative models are trained for classification and regression tasks using support vector machines. The whole learning scheme is evaluated by cross-validation applying an appropriate evaluation metric. Results are finally given by three data subsets, by classification and prediction success as well as by individual trait scores by means of confusion matrices, iterative accuracy plots and charts giving insights on feature space composition.

After having presented detailed results from personality modeling out of the perspective of exploiting different data sets with different inherent characteristics this chapter concludes on and discusses general tendencies across all differences in the data structure, i.e. results are now presented and discussed along a personality-centered perspective for personality classification and individual trait score prediction success. Finally, the chapter adds a comprehensive analysis of influencing factors and unexpected observations during processing.

This work combines interdisciplinary knowledge and experience from research fields of psychology, linguistics, audio-processing, machine learning, and computer science. This concluding chapter summarizes the main results as well as contributions of this work and gives a prospective for future work and application opportunities.