基于Wigner-Ville分布的移动机器人语音定位中时延估计方法

Time delay estimation algorithm based on Wigner-Ville distribution in voice localization of mobile robot

在移动机器人语音交互中,时延估计算法的精度是影响语音定位精度的关键因素。针对在强噪声情况下,基于时延估计原理进行语音定位的精度较低的问题,提出基于维格纳-威利分布(Wigner-Ville distribution,WVD)的一次与二次相关时延估计算法。在强随机噪声环境条件下,信号间的相关计算会加大侧峰的峰值,极大程度地干扰对主峰峰值的搜索,无法准确估计时延,进而影响语音定位效果。为了减小侧峰的影响,进一步对相关峰值加窗处理。但加窗虽减小了侧峰的干扰,却导致主峰宽度变大且变化趋势平坦,也会影响时延估计精度,因此,需选择希尔伯特变换对峰值锐化处理。仿真实验研究了各算法的估计情况与时延估计的均方根误差,实验结果表明,所提出算法相比其他几种算法对时延估计性能更高。此外,实际场景实验验证了该算法的抗噪声性能。

In the voice interaction for mobile robot,the time delay estimation algorithm is a key factor affecting the positioning accuracy of voice.Aiming at the problem of low precision of voice localization based on the time delay estimation in case of strong random noise,this paper proposes a Wigner-Ville distribution(WVD)based time delay estimation algorithm.Due to the strong random noise,the correlation calculation between signals will increase the peak value of side peak,which significantly interferes with searching the maximum from the main peak,and cannot accurately estimate the time delay.In order to reduce the influence of the side peak,the correlation values are processed by window function.Although the windowing reduces the interference of the side peaks,the width of the main peak becomes larger and the trend of the main peak is flat,which also affects the accuracy of the time delay estimation.As a result,the Hilbert transform is required to sharpen the peak value.The simulation experiment investigates the estimation of each algorithm and the root-mean-square-error(RMSE)of the time delay estimation,and the results demonstrate the proposed algorithm has a better performance than others.In addition,the real scene experiment further proves the anti-noise performance of the proposed method.