三阶累积量在语音信号端点检测中的应用

Third-order Cumulants Using in Voice Activity Detection

语音信号的三阶累积量通常不等于零,因而可以用来检测噪声环境中语音的起始点和终止点。提出一种采用累积量矩阵的最大奇异值来实现语音端点检测的方法,并引入一种自适应的实现方法。仿真实验表明,所提方法在很大程度上降低了传统端点检测方法的错误率。

Third-order cumulants of speech signals are not identically zero, so it could be used to detect the start point and end point of the voice under very noisy environments. The proposed method uses the maximum singular value of an cu- mulant matrix to distinguish between voiced parts of the speech signal and noise. Adaptive implementation is introduced, and the simulation results show that the proposed method reduced the error-ratio of the traditional algorithm.