Output details
11 - Computer Science and Informatics
University of Edinburgh
Head Motion Analysis and Synthesis over Different Tasks
<22> Originality: The first animated talking faces capable of learning and synthesising task-dependent head movements from data automatically. Employed articulatory features estimated from speech to improve the accuracy of audio-to-motion transformation.
Significance: Appropriate head movements besides realistic lip-synchronisation are crucial to make talking faces natural and believable. Existing studies employ short-time recorded data under strictly controlled conditions with specific emotions and personalities, which impede creating long-time animations of natural motions. The proposed method automatically learns typical head motions from long-time recordings of monologue/dialogue speech.
Rigour: Statistical analysis of head rotations in terms of speakers and tasks.