The phase transform prefilter based generalized cross-correlation is adopted widely in the estimation of the time-differences-of-arrival, since it performs very well under reverberant environment with low noise condition. However, it is not robust to high noise environments. In this paper, we propose to prefilter the signals in the wavelet domain with the aim of estimating the direction-of-arrival of speech source in the presence of low SNR. Simulation results show that our proposed algorithm can give rise to good direction-of-arrival performance compared with other well-known methods.