Automatic Lipreading in the Dutch Language

This thesis deals with many aspects of the bimodal speech processing research. It lays out the general framework of visually enhanced speech processing computer systems together with some insight in the human speech bimodal speech perception. There are three main contributions to the field of bimodal speech processing presented in the thesis. Firstly, a novel approach to visual feature extraction suitable for lipreading part of the speech processing system (Lip Geometry Estimation) is presented and described in full detail. Another new, powerful concept which is presented in this thesis is Person Independent Feature Space (PIFS), which is qualitatively analyzed on basis of real-life recorded material. The quantitative improvements obtained by using PIFS for lipreading applications are also presented here. The last major contribution of this research is the Delft University of Technology Audio-Visual Speech Corpus (DUTAVSC). This corpus has been extensively used throughout the research and provides a good starting point for future development of lipreading-capable speech processing systems.Information Technology and System