基于LPC高阶统计量语音活动检测算法的设计与实现

Study of Speech Activity Detective Algorithm Based on LPC Predictive Residual of Speech High Order Statistics

论文提出了一种基于LPC残差信号高阶统计量对加性噪声中的确定性信号及非高斯随机信号进行检测的方法,通过理论分析与实验结果表明,高阶统计量能有效地解决客观存在的非高斯及非线性问题,是现代信号分析与处理研究的重要内容,这种检测方法对加性噪声是否有色或是否为高斯分布并不敏感,而只要求噪声具有对称的概率密度函数。利用这种方法对确定性LPC残差信号进行检测的性能与已知波形时匹配滤波器的检测性能要好。利用这种方法还非常易于实现非高斯随机信号的检测。

This paper puts forward a kind of Speech activity detective algorithm based on LPC-Predietive-Residualof-Speech high order statistics to settle the random signal of non- Gausses and gauss signal, and the theoretical analysis and experiment results indicate that the high order statistics can resolve the objectively existed non- Gauss and nonlinear problem, is the important contents for modern signal analysis and processing study. This examination method is not sensitive to added noise, whether its color noise or Gauss distribution, so long as the noise has a symmetric density function. Making use of this kind of method to carry on the function examination is better than Match the filter. This method can easily realize non- Gauss examination of the random signal.