Real-time speech animation system

Abstract

We optimize the synthesis procedure of a videorealistic speech animation system [7] to achieve real-time speech animation synthesis. A synthesis rate must be high enough for real-time video streaming for speech animation systems to be viable in industry and deployed as applications for user machine interactions and real-time dialogue systems. In this thesis we apply various approaches to develop a parallel system that is capable of synthesizing real-time videos and is adaptable to various distributed computing architectures. After optimizing the synthesis algorithm, we integrate a videorealistic speech animation system, called Mary101, with a speech synthesizer, a speech recognizer, and a RTMP server to implement a web-based text to speech animation system.