Real-time Speech Driven Avatar with Constant Short Time Delay

It is shown that the perception of speech is inherently multimodal [16][22]. Auditory-visual speech recognition is more accurate than auditory only or visual only speech recognition [1][10]. Research shows that a synthetic talking face can help people understand the associated speech in noisy environments [16]. It also helps people react more positively in interactive services [20]. In some situations, reliable high bandwidth communication channel can hardly be guaranteed. Real-time speech driven avatar, which requires only very low bit rate channel to transmit information, provides an effective solution. This paper presents a method for realtime speech driven face animation with constant short latency.