机载环境下的语音端点检测  

Voice Endpoint Detection in Aviation Noise Circumstances

作　　者：刘静[1] 王建国[1] 孙文杰[2] 张帆[2] 

机构地区：[1]电子科技大学,成都610054 [2]中国西南电子技术研究所,成都610036

出　　处：《电讯技术》2008年第10期59-62,共4页Telecommunication Engineering

摘　　要：由于传统的语音端点检测方法仅以信号的短时能量、过零率等简单的单一特征作为判决特征参数,这些方法在实际应用中,很多都无法满足系统的需要。针对机载环境下的带噪语音进行研究,在传统方法的基础上作了两点改进:把短时能量和过零率结合为一个特征即短时能零积,并把其和短时自相关函数结合起来作为端点检测的特征参数。对于短时自相关的门限,不是利用传统的主副峰比值,而是利用达到副峰的时间间隔作为判决门限。通过对不同信噪比的机载环境下的带噪语音进行试验仿真,发现这种方法比单一的特征参数更有优越性,效果更明显。Traditional methods of speech endpoint detection based on single characteristic parameter can not satisfy the requirement of system in practical applications. In this paper, by analyzing the characteristic of aviation noise, two aspects of improvement are used to detect voice endpoint：（1）Temporal energy is combined with zero across rate as a characteristic parameter or energy - zero - product. Then this characteristic parameter is used as voice endpoint detection parameter together with short - time auto - correlation. （2） Time interval, from peak value to secondary peak value, is used as decision threshold of short - time auto - correlation. Simulation of noisy speech signals with different SNR under aviation noise circumstances is performed. The result shows that this method is prior to other methods based on single characteristic parameter.

关 键 词：语音识别 机载噪声 语音端点检测 能零积 短时自相关 判决门限 

分 类 号：TN912.3[电子电信—通信与信息系统]