Audio-Visual analysis of Multimedia documents for automatic topic identification.pdf

下载文档

1
0
约2.82万字
约 6页
2017-04-13 发布于江苏
举报
版权申诉
保障服务

Audio-Visual analysis of Multimedia documents for automatic topic identification.pdf

1、本文档共6页，可阅读全部内容。
2、有哪些信誉好的足球投注网站（book118）网站文档一经付费（服务费），不意味着购买了该文档的版权，仅供个人/单位学习、研究之用，不得用于商业用途，未经授权，严禁复制、发行、汇编、翻译或者网络传播等，侵权必究。
3、本站所有内容均由合作方或网友上传，本站不对文档的完整性、权威性及其观点立场正确性做任何保证或承诺！文档内容仅供研究参考，付费前请自行鉴别。如您付费，意味着您自己接受本站规则且自行承担风险，本站不退款、不进行额外附加服务；查看《如何避免下载的几个坑》。如果您已付费下载过本站文档，您可以点击这里二次下载。
4、如文档侵犯商业秘密、侵犯著作权、侵犯人身权等，请点击“版权申诉”（推荐），也可以打举报电话：400-050-0827(电话支持时间：9:00-18:30)。

Audio-Visual analysis of Multimedia documents for automatic topic identification

AUDIO-VISUAL ANALYSIS OF MULTIMEDIA DOCUMENTS FOR AUTOMATIC TOPIC IDENTIFICATION Uri Iurgel, Steffen Werner, Andreas Kosmala, Frank Wallhoff Gerhard-Mercator-University of Duisburg Department of Computer Science iurgel,werner,kosmala,wallhoff @fb9-ti.uni-duisburg.de Gerhard Rigoll Munich University of Technology Institute for Human-Machine Communication rigoll@ei.tum.de ABSTRACT This paper presents a system that shall automatically scan multimedia data like TV or radio broadcasts for the pres- ence of specific topics and, whenever topics of users’ inter- ests are detected, alert the related user. Our current work on the three main modules of the system will be shown. (1) The speech recognition system (with 18.7 % WER) is already among the most advanced German broadcast speech recognition systems. (2) The new and innovative topic identification approach, which is especially designed to work on the output of a speech recognizer, is compared to a standard text based approach. (3) The topic segmentation module has a good perfor- mance detecting real topic boundaries, not just scene cuts or speaker turns. KEY WORDS Audio and Video, Multimedia, Speech Processing, Topic Segmentation, Topic Identification 1 Introduction Motivation Newspapers, magazines, radio, television, world wide web - information is of strategic importance for business and governmental agencies as well as for citizens. The ex- ponential evolution of multimedia makes it difficult to overview the opulence of information, and that’s why im- portant pieces of information have to be filtered and pro- cessed automatically. Nowadays, information is mainly obtained by manu- ally analyzing (reading, listening and watching) large au- dio and video databases and current broadcast multimedia sources (such as broadcast TV, radio or Internet streams). After having assigned topics to the incoming news and sto- ries, only the items of interest or items regarding a specific request will be selected and further proc