中文摘要：

为了增强相位变换加权广义互相关方法（GCC-PHAT）这一常用时延估计方法对噪声的鲁棒性,提出了一种基于频率自适应选择的改进算法.该算法利用较短的语音数据（32ms）估计出每个频率的协方差矩阵,从而计算出每个频率的信噪比（SNR）,然后自适应地选择SNR较大的频率用于时延估计.仿真实验结果表明,不论是在中等混响（混响时间T60=0.3s）还是在强混响（T60=0.6s）条件下,相对于GCC-PHAT方法,该算法对噪声的鲁棒性更强.

英文摘要：

The generalized cross correlation approach with phase transform weighting（GCC-PHAT） is a popular time delay estimation（TDE） algorithm.To enhance the robustness of the GCC-PHAT against noise,an improved TDE algorithm based on adaptive frequency selection is proposed.The signal-to-noise ratio（SNR） of each frequency is calculated from the covariance matrix estimated from the short segments（32ms） of speech,and then the frequencies with high SNR are adaptively selected to estimate the time delay.Simulation results show that the proposed method is more robust against noise compared to the traditional GCC-PHAT method under both medium（reverberation time T60=0.3s） and high reverberation（T60=0.6s） conditions.