标题: | 电视新闻语音检索之研究 The Study of Spoken Document Retrieval on TV news |
作者: | 蔡富评 Fu-Ping Tsai 傅心家 Hsin-Chia Fu 资讯科学与工程研究所 |
关键字: | 语音资讯检索;语者分段;语者分群;语者识别;Speech Information Retrieval;BIC Segmentation;BIC Clustering;Speaker Identification |
公开日期: | 2005 |
摘要: | 语音资讯检索主要是研究如何对大量的多媒体资讯(如广播新闻),利用语音辨识技术,以自动的方式对于其内含的语音资讯建立起全文索引与检索的机制。本篇论文主旨在针对台湾广播新闻,在建立语音检索的机制之前,需要针对电视新闻节目建立起自动新闻分析的系统,以侦测出新闻节目中主播的位置并切割新闻故事的问题作探讨研究。近来许多新闻节目中主播音段常有明显的背景音乐,为了正确的侦测出没有背景音乐的主播音段,论文中提出结合BIC语者分段与分群以及语者识别的技术来侦测新闻中没有背景音乐的主播音段。我们以台湾有线东森新闻台的新闻节目进行主播侦测的实验,验证所提的方法能正确侦测出没有背景音乐的主播音段,论文最后更进一歩实作语音音节辨识并且成功建立起以音节为索引特征之电视新闻语音检索系统。 This thesis mainly describes broadcast news retrieval system for Mandarin Chinese. First, we need to construct automatically news analysis system to detect anchor segments in news program. Recently, we observed some anchor segments that have background music in many news programs. In order to correctly detect anchor segments without background music, we propose a method based on technologies such as BIC-Segmentation, BIC-Clustering and GMM-based speaker identification for TV news anchor detection. The experiment corpus is collected from daily news on ETT news program and the experiment result is good. Moreover, we integrate the proposed method and implement syllable-level indexing feature news spoken document retrieval system on TV news successfully. |
URI: | http://140.113.39.130/cdrfb3/record/nctu/#GT009317577 http://hdl.handle.net/11536/78786 |
显示于类别: | Thesis |
文件中的档案:
If it is a zip file, please download the file and unzip it, then open index.html in a browser to view the full text content.