一种基于近邻传播聚类的语音端点检测方法  

Voice activity detection based on affinity propagation clustering

作　　者：林琴 涂铮铮[2] 王庆伟 郭玉堂 LIN Qin;TU Zhengzheng;WANG Qingwei;GUO Yutang(School of Computer Science Technology, Hefei Normal University, Hefei 230601, China;School of Computer Science and Technology, Anhui University, Hefei 230601, China;Anhui Weitai Intelligent Technology Co. Ltd, Hefei 230088,China)

机构地区：[1]合肥师范学院计算机学院,安徽合肥230601 [2]安徽大学计算机科学与技术学院,安徽合肥230601 [3]安徽威泰智能科技有限公司,安徽合肥230088

出　　处：《安徽大学学报（自然科学版）》2019年第3期27-32,共6页Journal of Anhui University(Natural Science Edition)

基　　金：国家自然科学基金青年基金资助项目(61602006);安徽省高校自然科学研究重点项目(KJ2013A217;KJ2017A934)

摘　　要：为提高语音端点检测在低信噪比情况下的准确性,提出一种基于近邻传播聚类的语音端点检测算法.首先采用能量语音端点检测去除静音段;然后利用近邻传播聚类自动获取类别数的优点,有效地将语音细分为无语义语音和静音段、远场噪声段等各种类别;最后结合后处理方法,对语音端点做进一步过滤处理.实验结果表明:该算法在低信噪比的情况下,与传统的能量语音端点检测相比,其有效语音检测的漏警率相对下降13%,虚警率相对下降14%;在实际应用中,如声纹确认和声音检测等,与经典算法相比,该算法检测的准确率与效率等性能得到了显著提升.To improve the accuracy of voice activity detection under low signal-to-noise ratio conditions, a novel method based on affinity propagation clustering was proposed in the paper. Firstly, the mute segment was removed by energy voice activity detection;then, the advantages of the number of categories were automatically obtained by using affinity propagation clustering, and the speech was effectively divided into various categories such as non-semantic speech, silent segments and far-field noise segments;finally, combined with the post-processing method, the voice activity detection was further filtered. Compared with traditional energy voice activity detection methods, the results of affinity propagation clustering method had a relative drop of 13% in missing alarm rate for effective speech detection and the false alarm rate had decreased by 14%. In practical applications, such as voiceprint confirmation and sound detection, compared with the classical algorithm, the performance of the algorithm in terms of accuracy and efficiency of detection was significantly improved.

关 键 词：语音端点检测 近邻传播聚类 漏警率 虚警率 

分 类 号：TN912.34[电子电信—通信与信息系统]