###
DOI:
计算机系统应用英文版:2014,23(10):40-45
本文二维码信息
码上扫一扫!
广播电视新闻中的主持人跟踪系统
(1.电子科技大学 电子工程学院, 成都 611731;2.湖南人民广播电台 技术部, 长沙 410007)
Anchor Speakers Tracking in Broadcast News
(1.School of Engineering, University of Electronic Science and Technology of China, Chengdu 611731, China;2.Technology Department, Hunan Broadcasting System, Changsha 410007, China)
摘要
图/表
参考文献
相似文献
本文已被:浏览 1508次   下载 2725
Received:February 10, 2014    Revised:March 17, 2014
中文摘要: 针对广播电视新闻节目中的主持人跟踪问题,提出了一种将说话人分割聚类和说话人确认有效结合的算法,并根据该算法设计了一套主持人跟踪系统. 该系统首先利用音频活动检测算法去除新闻音频资料中的静音段,然后说话人分割聚类算法将多说话人语音段分成若干单一话者语段,最后通过基于GMM-UBM的说话人确认算法辨认每段单一话者语段的话者身份是否为目标主持人. 此外,分析了T-Norm对系统性能的影响. 以中央电视台《新闻联播》为评测数据集,实验结果表明,该算法取得了良好的效果,跟踪系统的查准率(Precision)和查全率(Recall)分别为93.03%和84.34%.
Abstract:To address the problem of anchor speakers tracking in broadcast news, an algorithm that integrated effectively speaker segmentation and clustering and speaker verification is presented. An anchor speakers tracking system based on the proposed algorithm is also designed. The system firstly uses audio activity detection algorithm in order to remove silence segments. Secondly, a speaker segmentation and clustering process converts multi-speaker speech waveform into several speaker homogenous segments. Finally, speaker verification approach based on GMM-UBM is dedicated to decide whether a segment belongs to one targeted anchor speaker. Furthermore, the impact of T-norm on system performance is also analyzed. Experiments on CCTV Mandarin Broadcast News demonstrate the effectiveness of the proposed algorithm. The tracking system achieves precision and recall at 93.03% and 84.34% respectively.
文章编号:     中图分类号:    文献标志码:
基金项目:
引用文本:
汪洋,甘涛,向军.广播电视新闻中的主持人跟踪系统.计算机系统应用,2014,23(10):40-45
WANG Yang,GAN Tao,XIANG Jun.Anchor Speakers Tracking in Broadcast News.COMPUTER SYSTEMS APPLICATIONS,2014,23(10):40-45