ON TIME WARPING ALGORITHMS FOR SPEECH RECOGNITION.

In this paper, two simplified time warping algorithms are suggested. A comparison between these two algorithms and two well known algorithms, often discussed in the literature, is presented. The two algorithms most often used in the literature are the Linear Time Warping (LTW) and the Dynamic Programming Time Warping (DPTW). The two simplified methods that have been developed and described in this paper are based on the fact that speech utterances can be sectioned into three types of signals: Voice (V), Unvoice (U) and Silent (S). The proposed two methods provide a compromise between computation load and warping efficiency. A comparison of the four time warping methods was performed using a text dependent speaker verification experiment.