带噪语音端点检测方法研究  

Study on noisy speech endpoint detection method

作　　者：朴春俊[1] 马静霞[1] 徐鹏[1] 

机构地区：[1]燕山大学电气工程学院,河北秦皇岛066004

出　　处：《计算机应用》2006年第11期2685-2686,2690,共3页journal of Computer Applications

基　　金：河北省研究性发展计划项目(04213537)

摘　　要：影响语音识别性能的一个关键因素是端点检测的准确性。实际应用中信噪比较低,使得某些高信噪比下性能好的检测算法不能有效工作,影响系统的识别率。提出了一种基于时频方差和的语音端点检测算法。实验证明该算法能够在低信噪比的情况下,准确地检测出语音信号。通过对三种不同的端点检测算法的比较,发现基于时频方差和的端点检测算法的端点检测的准确率较高。A major factor influencing the capability of speech recognition systems is the accuracy of endpoint detection, Some good detection algorithms cannot take effect in practical use because small SNR reduces the recognition rate. An algorithm for speech endpoint detection based on time-frequency-variance-summation was proposed. The experiments prove that it has better detection capability with low SNR compared with other three endpoint detection algorithms.

关 键 词：语音识别 端点检测 时频方差和 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术] TN912.34[自动化与计算机技术—计算机科学与技术]