低信噪比环境下语音端点检测改进方法-声学技术.PDFVIP

第37 卷第4 期 声 学 技 术 Vol.37, No.4 2018 年8 月 Technical Acoustics Aug., 2018 低信噪比环境下语音端点检测改进方法 王 瑶，曾庆宁，龙 超，谢先明，毛 维 (桂林电子科技大学认知无线电与信息处理教育部重点实验室，广西桂林541004) 摘要 ：针对语音端点检测在低信噪比环境下普遍存在检测性能急剧下降的问题，提出一种将调制域(时间-频率域)谱 减法和自相关函数相结合的语音端点检测算法。该算法首先利用调制域谱减法较好的消噪能力来提高含噪语音的信 噪比；然后根据语音和噪声的自相关函数的主峰最大值和次大值之比差异较大的特性，结合基于对数能量和自相关 函数的端点检测方法对消噪后的语音进行端点检测。实验结果表明，该算法在低信噪比的环境下能取得较好的端点 检测效果，并具有较好的稳健性。 关键词：低信噪比；调制域；自相关函数；对数能量；端点检测 中图分类号：TN912.35 文献标识码 ：A 文章编号 ：1000-3630(2018)-05-0457-11 DOI 编码 ：10.16300/ki.1000-3630.2018.05.010 An improved speech endpoint detection method under low SNR WANG Yao, ZENG Qing-ning, LONG Chao, XIE Xian-ming, MAO Wei (Key Laboratory of Cognitive Radio and Information Processing of Ministry of Education, Guilin University of Electronic Technology, Guilin 541004, Guangxi, China) Abstract: In this paper, a new approach combining the spectral subtraction in modulation (time-frequency)domain and the post processing for the autocorrelation functions of signal and noise is proposed to improve the performance of speech endpoint detection in low SNR environment. Firstly, the modified spectral subtraction used in modulation do- main reduces the noise to increase SNR. Then, according to the feature we figure out that a quite difference in the ratio of maximum to secondary value of the peak of autocorrelation function exists between speech and noise, a method based on logarithmic energy and autocorrelation function is used for endpoint detection of the sp