This article aims at the demand of voice call control for communication equipment, synthetically analyze the main methods and performance of speech endpoint detection. Propose a speech endpoint detection algorithm based on spectral entropy. Introduce the algorithm of speech endpoint detection and calculate spectral entropy of the speech signal, through setting up the background noise model, the threshold of the spectral entropy is dynamically adjusted to realize the function of speech endpoint detection. Simulation results show that the algorithm not only has good performance of speech endpoint detection under low SNR, but also has good adaptive ability in complex background noise environment.