Time-Normalization Techniques for Speaker-Independent Isolated Word Recognition

In this paper, we investigate various timenormalization techniques that are useful in the context of speaker-independent isolated word recognition. At the lowest level, we make use of LPC coefficients as the features to be nomalized. We discuss the various methods by which we can normalize these features. To begin with, we arrive at a typical number of frames associated with a word. Then, we normalize all the training and test data to this number of frames. Initial results bring out the point that normalization techniques help in reducing the number of patterns with which the unknown has to be compared