Journals Proceedings

Abstract

In this paper we present results of visual speech recognition using geometric and appearance model. For each of these methods Principal Component Analysis (PCA) is used to compute feature vectors. In this paper visual speech recognition is used for isolated digit utterance. For determination of visual features lip tracking is important. We used localized active contour method for lip tracking which was earlier used for biomedical application. PCA distance methods are used for feature vector comparison. Appearance method is compared with geometrical method for feature extraction. Appearance method is computationally efficient and giving better result as compare to geometrical method. Using appearance method of feature extraction, effect of isolation between two digit utterances is tested.