一种时频结合的抗噪性端点检测算法  

Anti-noise Endpoint Detection Algorithm Based on Time-frequency Joint

作　　者：李灵光[1] 

机构地区：[1]北京工业大学软件学院,北京100022

出　　处：《计算机与现代化》2011年第8期29-31,共3页Computer and Modernization

摘　　要：端点检测是语音识别中的一项关键技术,检测的准确性对语音识别的性能有很大的影响。本文提出一种时频结合的基于子带谱熵和短时能量的端点检测方法,和常用的双门限法、谱熵法相比在低信噪比下检测的准确率有很大提高。该方法具有抗噪性、实时性等特点,适用于实际的语音识别应用。Endpoint detection is an essential technology in a speech recognition system,its accuracy has a very tremendous influence on the speech recognition performance.This paper proposes a band-partitioning spectral entropy and short-time energy based endpoint detection algorithm based on time-frequency joint.Compared with the two-door method and spectral entropy,its precision is greatly improved.This method has characteristics for noise immunity and real-time,and can be applied to practical application of speech recognition.

关 键 词：端点检测 子带谱熵 时频结合 短时能量 信噪比 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]