With the development of multimedia and web technologies, ever-increasing multimedia collections are available, e.g. broadcast news, meetings and lectures. Given the vast amount of multimedia data, automatic approaches for multimedia processing are urgently in demand, especially for automatic indexing, summarization, retrieval, visualization, and organization technologies. Among these technologies, automatic story (or topic) segmentation is an important precursor since other tasks usually assume the presense of individual topical documents. Story segmentation is such a tasks that divides a stream of text, speech or video into topically homegeneous blocks, known as stories. Specifically for broadcast news (BN), a popular media repository, the objective is to segment continuous audio/video streams into distinct news stories, each addressing a central topic.
Automatic story segmentation was the research topic for my master project. We proposed various approaches to solve this problem by adopting TextTiling approach, analyzing prosodic features from audio and integrating multi-modal features, and several academic paper have been published.
Lei Xie, Chenglin Xu, Prosody-based Sentence Boundary Detection in Chinese Broadcast News , ISCSLP, 2012, Hong Kong, China.,
Broadcast News Story Segmentation Using Conditional Random Fields and Multi-modal Features, IEICE Transaction on Information and Systems, 2012., Lei Xie, Mimi Lu, Bin Ma, Eng Siong Chng, Haizhou Li,
Phoneme Lattice based TextTiling towards Multilingual Story Segmentation , INTERSPEECH, 2010, Portland, U.S.A., Lei Xie, Bin Ma, Eng Siong Chng, Haizhou Li,
Modeling Broadcast News Prosody Using Conditional Random Fields for Story Segmentation , IAPSIPA ASC, 2010, Singapore., Lei Xie, Bin Ma, Eng Siong Chng, Haizhou Li,