Word Spotting Using Common Vector Approach

Abstract

Common vector approach (CVA) is a subspace method and it aims to find a unique vector which contains the common features for each class. CVA was successfully applied to pattern recognition experiments like isolated word recognition, image recognition and multi-class cases. It is aimed here to set out a novel application of CVA, word spotting in continuous speech. Two different recordings containing ten keywords were used for training and testing. A Hundred percent successful recognition was achieved with the aid of a pre-calculated decision threshold. However, the aim was to develop an algorithm independent of databases so a method was used to calculate threshold from training set. Again a hundred percent recognition was obtained on test set. The next step is to devise a totally autonomous recognition system and obtain more experimental data on universal databases.