【摘要】 根据语音信号的短段循环平稳(CycloStationary,CS)特征,该文提出了一种应用于复杂背景噪声条件下的基于高阶循环累积量的改进型VAD(Voice Activity Detection)算法,算法采用MA(Moving Average)模型对语音信号建模,并选择平均幅度差(Average Magnitude Difference Function,AMDF)的方法来估算循环频率以降低算法复杂度。经VolP(Voice over Internet Protocol)平台测试,算法对高斯(白色或有色)噪声以及其它平稳噪声自适应能力强、检测性能突出,且恢复后语音质量损失较小,对于非对称噪声也具备可检测能力。更多还原

【Abstract】 On the basis of cyclostationary feature of speech in short segment, an improved Voice Activity Detection (VAD) algorithm based on higher-order cyclic cumulant is proposed in this paper, which can be applied in complex noise background. In this algorithm, modeling speech signal under MA(Moving Average)model is adopted and estimating cyclic frequency by AMDF(Average Magnitude Difference Function) is chosen to decrease computing complexity as well. Extensive testing on VoIP platform reveals that the algorithm owns good adaptive classification capability in steady noise, such as Gauss (whited and colored) noise, while with a little loss in speech quality, even when it is applied in asymmetric noise conditions.更多还原