A New and Efficient Alignment Technique by Cosine Distance

In this paper we describe a new technique to measure the similarity or distance between time series. We have called it, Alignment Technique by Cosine Distance (ATCD). Important features about the technique are that it requires neither a-priori knowledgement of the time series nor training stages. ATCD is based on cosine distance and least squares, and requires as a parameter the dimension of two support vectors. When we consider high dimensionality on these vectors, ATCD achieves its best performance providing the smallest measure of similarity (distance) as possible. ATCD can be used on applications of medical signal processing, audio and speech recognition, among others. We proved ATCD′s efficiency on an isolated-words speech recognition system by comparing ATCD against Dynamic Time Warping.