一种新的语音和噪声活动检测算法及其在手机双麦克风消噪系统中的应用

针对现有双通道语音活动检测(Voice Activity Detection, VAD)算法依赖于固定阈值难以在多种噪声环境下准确地检测语音和噪声，应用于手机消噪系统会造成语音失真或噪声消除不好等问题，该文提出一种基于神经网络的VAD算法，该算法以分频带能量差和归一化互通道相关为特征，采用神经网络对语音和噪声进行分类。在此基础上，将神经网络VAD与基于互通道信号功率比值的VAD相结合，提出一种新的适用于手机消噪系统的语音和噪声活动检测算法分别对语音和噪声进行检测，并以此进行噪声抑制处理，减少了消噪系统因VAD误判而造成的性能下降。实验结果表明，该处理方法在抑制背景噪声和减少语音失真等方面优于现有的消噪算法，对于方向性语音干扰也有很好的抑制效果。

A New Voice and Noise Activity Detection Algorithm and Its Applicationto Dual Microphone Noise Suppression System for Handset

Existing dual microphone Voice Activity Detection (VAD) algorithms use normally a fixed threshold. The fixed threshold can not provide an accurate VAD under various noise environments. In such case, it causes voice quality degradation, particularly in handset applications. This paper proposes a new VAD algorithm based on Neural Network (NN). Both sub-band power level difference and inter-microphone cross correlation are used as features. Then the NN based VAD is combined with the method of inter-microphone signal power ratio to get a new voice and noise activity detection algorithm. Furthermore, the algorithm is used into noise suppression in handset to avoid performance degradation caused by VAD misjudgment. Experimental results show that the proposed method provides better noise suppression performance and lower speech distortion compared to the existing method.