Author : Guru keerthana Gaddam 1
Date of Publication :7th November 2016
Abstract: With large volumes of multimedia data and speech recordings available over internet,there is need to efficiently process these data so that users can quickly review important information. So the research is mainly focused on automatic processing of transcripts.In past decades,lot of methods have been proposed for summarization of text. The solution for speech summarization is to transliterate the spoken documents to texts, and apply some well-defined text summarization methods. When these methodologies are applied to spoken documents, they doesn’t work good for text processing. Shih-Hung Liu et al. performed speech summarization by combiningClarity Measure andRelevance Language Modelling(RLM). Clarity measure is used for important sentence selection, which helps to identify the individual sentences which reflect the main theme of the document. The experimental evidence of this model indicated that the various formulations instantiated are better than few existing methods for extractive speech summarization. The sentence-level clarity measure in combination with RLM indeed benefits speech summarization significantly. The limitations observed in this model are that, it is purely term based and doesn’t consider concept relevance. So this project aims at proposing the use ofCorpus or Knowledge base for Extractive speech summarization,where a subset of sentences will be selected to cover as many important concepts as possible.
Reference :