Static vs. dynamic modeling of human nonverbal behavior from multiple cues and modalities