Extracting and analysing co-speech head gestures from motion-capture data

This paper reports on a method developed for extracting and analyzing head gestures taken from motion capture data of spontaneous dialogue in Swedish. The head gestures were extracted automatically and then manually classified using a 3D player which displays time-synced audio and 3D point data of the motion capture markers together with animated characters. Prosodic features were extracted from syllables co-occurring with a subset of the classified gestures. The head gestures show considerable variation in temporal synchronization with the syllables, while the syllables generally show greater intensity, higher F0, and greater F0 range when compared to the mean across the entire dialogue.