In order to improve the accuracy and speed of time-delay estimation(TDE) in the case that the signals are non-stationary and contain a large number of sampling points, a new method for generalized cross-correlation TDE based on signal chopping-based CEEMD (Complementary Ensemble Empirical Mode Decomposition, CEEMD) de-noising method is proposed. In this method, firstly the original large-signal is devided into several smaller sections of equal, then each section is de-noised using CEEMD de-noising method, all the smaller sections are combined to obtain the de-noised version of the original signal, finally the time delay value is estimated according to the cross-correlation peak of the de-noised signals. For the selection of the section number and the number of added noise in the process of TDE, the specific range of these two parameters is given by exploring the effect of these two parameters on de-noising precision, calculation time and other performance of the simulated people walking signal. Finally, this method is applied to the TDE in acoustic source localization system, results show that, this method can still rapidly and accurately estimate time delay value when the signals contain a large number of sampling points and the SNR is low.