基于二次相关的语音信号时延估计改进算法

目前语音信号的时延估计研究,大部分采用的是广义互相关算法。然而,广义互相关时延估计算法易受噪声和混响环境影响。为此,本文提出了一种基于二次相关的语音信号时延估计改进算法,该算法对语音信号进行二次互相关运算,并结合Hilbert变换,对二次互相关峰值进行进一步的锐化处理,使得反映时延的峰值点检测更为准确。实验结果表明,改进的时延估计方法在非平稳的语音信号中能够有效地抑制噪声干扰,且在不同混响条件下时延估计具有更好的性能。

A proposed time delay estimation in speech signal based on second correlation

The generalized cross correlation algorithm has been widely used in the time delay estimation of speech signals. However, the performance of the algorithm can be easily influenced by the presence of noises. A new algorithm based on the second correlation method has been proposed in this paper to estimate the time delay of speech signals. The algorithm calculates the second correlation of speech signals, and then sharpens the second correlation peak values by combining with the Hilbert transform. As a result, the detection of the peak point representing time delay becomes more precise. The experiments show that the second correlation method can suppress noise interference more effectively. And even in the different circumstances of reverberation, the proposed algorithm can obtain higher delay estimation accuracy and better stability for non-stationary speech signals.