论文部分内容阅读
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
The traditional speech endpoint detection method uses the simple features of signals such as short-time energy and zero-crossing rate as the decision feature parameters. In practice, these methods can not meet the needs of the system, especially when the signal-to-noise ratio is relatively low. In this paper, the cepstrum feature of speech signal is used as the basis for judging whether the sampled signal frame is speech signal or not, and the cepstrum distance measurement method and cyclic neural network method are proposed. Through the experiment of broadband noise-white noise interference and a special noise-car noise situation, it is found that the endpoint detection method of speech signal with cepstrum characteristic parameters has the unparalleled superiority of traditional energy method in noise environment, and is more suitable In practical application