There is increasing socio-emotional competence in present-day human-machine, human-robot, and human-agent communication allowing to build virtual companions beyond the uncanny valley in the near future. Reaching from emotion to personality and more general from a person's state to her traits, such information can be derived from human verbal and non-verbal behaviour.
It is the aim of this talk to introduce the computational assessment of a variety of subject characteristics focused on audio, but also audiovisual and more general multimodal input analysis. Subsequent to a discussion of the signal front end side including enhancement of the signal of interest and feature-space brute-forcing, the computational back-end side will be touched. Here, memory enhanced network structures are presented. Before providing a résumé, a number of practical applications and "real-world" experiments are shown, including research Challenges organised by the presenter.