The Dynamic Time Warping(DTW) algorithm is a nonlinear warping algorithm combined with dynamic time warping technique and distance measurement computing method. It has an important application in speech recognition based on template matching. This paper proposes an improved efficient dynamic time warping algorithm, which can effectively speed up the path searching in speech recognition. Based on Matlab, the hidden Markov algorithm, efficient dynamic time warping algorithm and improved dynamic time warping algorithm speech recognition systems are implemented. Simulation experiments based on above algorithms are performed. The experimental results show that the training speed of improved efficient dynamic time warping algorithm is much faster than that of efficient dynamic time warping algorithm and hidden Markov algorithm, but the recognition rate of decline based on improved efficient dynamic time warping algorithm is very low. For small vocabulary continuous speech recognition, the recognition rate is 97.56% for efficient dynamic time warping algorithm, 97.14% for hidden Markov recognition algorithm, and 96.43% for improved efficient dynamic time warping algorithm respectively.