短时TEO能量在带噪语音端点检测中的应用  

Application of short-time TEO energy in noisy speech endpoint detection

作　　者：李杰[1] 周萍[2] 杜志然[1] 

机构地区：[1]桂林电子科技大学计算机科学与工程学院,广西桂林541004 [2]桂林电子科技大学电子工程与自动化学院,广西桂林541004

出　　处：《计算机工程与应用》2013年第12期144-147,共4页Computer Engineering and Applications

基　　金：国家自然科学基金(No.60961002)

摘　　要：语音端点检测是语音识别系统的一个重要组成部分,特别是在噪声环境下,其准确性直接影响到语音识别系统的计算复杂度和识别性能。提出了一种在噪声环境下基于短时TEO能量的语音信号端点检测方法,采用了双门限-三态转换判决机制以保证算法在噪声环境下的端点检测准确性和对信号绝对幅度变化的稳健性。实验结果表明,与传统的短时能量法和谱熵法相比,该算法在低信噪比情况下具有更好的端点检测能力,显示了算法的优越性。Speech endpoint detection is a crucial component in speech recognition system, especially in noisy environment. Its accu- racy affects the computational complexity and the recognition performance of the speech recognition system. This paper proposes an endpoint detection of speech signals based on short-time TEO energy in noisy environment. It uses a three-state transition and judgment mechanism based on double thresholds, which ensures the accuracy in noisy environment and the robustness to changes in absolute levels. Compared with same traditional algorithms such as short-time energy and spectral entropy, experiment results show this algorithm has better detection capability in low signal to noise ratio environments and takes on more advantages.

关 键 词：TEAGER能量算子 端点检测 语音识别 噪声 

分 类 号：TP391.42[自动化与计算机技术—计算机应用技术]