Automatic audio driven animation of non-verbal actions

While speech driven animation for lip-synching and facial expression synthesis from speech has previously received much attention, there is no previous work on generating non-verbal actions such as laughing and crying automatically from an audio signal. In this article initial results on a system designed to address this issue are presented. 3D facial data is recorded for a participant making different actions-i.e. laughing, crying, yawning and sneezing-using a Qualysis (Sweden) optical motion-capture system while simultaneously recording audio data. 30 retro-reflective markers were placed on the participant´s face to capture movement. Using this data, an analysis and synthesis machine was then trained consisting of a dual-input Hidden Markov Model (HMM) and a trellis search algorithm which converts HMM visual states and new input audio into new 3D motion-capture data.