一种噪声环境下连续语音识别的快速端点检测算法  

A Fast Endpoint Detection Algorithm for Continuous Speech Recogni tion in Noisy Background

作　　者：崔冬青[1] 李治柱[1] 吴亚栋[1] 

机构地区：[1]上海交通大学计算机系,上海200030

出　　处：《计算机工程与应用》2003年第23期95-97,138,共4页Computer Engineering and Applications

摘　　要：根据汉语语音的特点,该算法利用幅度及功率谱对语音端点进行检测,有效地消除了背景噪声及DC分量的干扰。算法采用实际语音采样进行分析,试验结果表明此算法不仅能有效地标识出语音的起始及终止点,并且还具有相当高的运算效率。According to the character of Chinese speech,this algorithm effectively elimilates the deviation caused by background noise and DC-offset in locating the endpoints of speeches in continuous speech recognition by utilizing amplitude and power spectra.Some real audio inputs are taken as examples to evaluate the performance of this algorithm.The result shows that this Algorithm can effectively mark the endpoints of each audio input.

关 键 词：功率谱 端点检测算法 DC分量 汉语 语音识别 背景噪声 

分 类 号：TN912.34[电子电信—通信与信息系统] TP301.6[电子电信—信息与通信工程]