How to accurately measure the speed of high-speed train has always been a hot issue in the research on high-speed railway. This paper introduces generalized cross-correlation time delay estimation algorithms to analyze dual-station recorded waves of ground vibrations induced by trains, and calculates train speeds. We capture waves of ground vibrations induced by trains along the high-speed railway between Yunnanyi station and Xiangyun station in the Guangtong—Dali railway section with seismographs. Then we analyze waves recorded in a pair of seismographs with three generalized cross-correlation time delay estimation algorithms, or PHAT weighting, SCOT weighting and ROTH weighting, and with basic cross-correlation time delay estimation algorithm respectively. The results of analyses show that compared to the other three algorithms the PHAT weighting effectively sharpens the maximum of cross-correlation function and the obtained train speeds based on this algorithm are more stable and accurate. Besides, PHAT weighting can avoid to a certain extent the problems met in the general methods of train speed measuring, such as the poor precision of instrument, the difficulties in instrument installation, etc., and can provide accurate train speed measurement for many areas where a prior information cannot be obtained. We calculate the train speed results of the three-component waveforms separately to determine which components are more accurate.When there are multiple railways, three-component autocorrelation waveform of train can be used to estimate the distance between the train and stations. For the obvious error of the train speed, we choose two wrong data to analyze the cause of the error, which is valuable in application. With the development of generalized cross-correlation technology, the calculation of time delay becomes more and more accurate.We will continue to follow the issues related to high-speed rail research and producing more valuable conclusions.