Kohonen clustering networks for use in Arabic word recognition system

Speech is the future mean of communication between man and machines. In this paper, we propose a speaker-independent isolated Arabic word recognition system, based on neural network. The speech signal is usually segmented into a sequence of frames in most of the speech processing techniques. These frames may overlap one another with a specific spacing. At each frame the extracted features form a feature vector. Then, an utterance can be represented by a sequence of feature vectors. This feature vector sequence is considered as speech pattern. The speech recognition is to classify the speech pattern and to identify the spoken words corresponding to the speech patterns. In the present study we use the Kohonen Clustering Networks algorithm to classify the speech pattern.