【摘要】 传统的语音端点检测方法以信号的短时能量、过零率等简单特征作为判决特征参数．这些方法在实际应用中，尤其当信号信噪比比较低时，无法满足系统的需要．文中利用语音信号的倒谱特征作为判决抽样信号帧是否为语音信号的依据，并提出了倒谱距离测量法和循环神经网络法．通过对宽带噪声－白噪声干扰情况和一种特殊噪声——汽车噪声情况的实验，发现倒谱特征参数的语音信号端点检测方法在噪声环境下具有传统的能量方法无法比拟的优越性，更适合于实际应用更多还原

【Abstract】 Most practical automatic speech recognition(ASR) systems must work with a small signal noise ratio(SNR), and the conventional speech detection methods based on some simple features such as energy cannot work well in the noisy environments. In this paper, cepstrum was used as the feature to detect the voice activity. Two algorithms for endpoint detection of noisy speech signal were proposed. The first one takes the cepstral distances as the decision thresholds instead of short time energy. The second approach takes advantages of recurrent neural networks. The experiments show that the high accurate rates can be obtained in the noisy cases.更多还原