项目名称: 基于结构建模的语音理解及应用研究
项目编号: No.61300197
项目类型: 青年科学基金项目
立项/批准年度: 2014
项目学科: 自动化技术、计算机技术
项目作者: 张剑
作者单位: 东莞理工学院
项目金额: 20万元
中文摘要: 近年来发展的语音理解是建立在语音识别和自然语言理解之上的新兴研究领域。本项目主要研究基于结构建模的语音理解与摘要技术,以能够帮助人们从语音文档中抽取出隐含的结构信息,更快速准确地理解语音文档,已成为本领域研究热点和前沿。由于语音识别技术的不完善,所生成文本含有不准确之处,对传统的基于文本结构建模是极大的挑战。针对此问题,本项目将在特征提取、模型算法及应用方面,在现有工作基础上进行深入研究:分析深层次语音文档结构,寻找对抽取语音文档结构帮助更大的新特征;应用不同的机器学习算法,提高抽取语音文档结构的性能,并运用抽取的结构信息,改善语音理解与摘要抽取的性能;将新算法应用于演讲语音领域。通过本项目的研究,一方面能够建立更有效的针对深层次语音文档结构建模的新算法,提高抽取语音摘要的性能与效率;另一方面基于结构建模的语音理解为海量语音文档管理与安全提供更好的理论支持与实用途径。
中文关键词: 结构建模;语音文档结构;语音理解;语音摘要;抽取式摘要
英文摘要: Speech Understanding (SU) is a young and under-exploited research field based on Automatic Speech Recognition (ASR) and Natural Language Understanding (NLU). In the proposal, our research work focuses on how to apply structural modeling technology for speech understanding and summarization, which can help users understand speech documents faster and more exactly by hidden structure information extracted from speech documents. More and more researchers take their interests in this topic. Considering that speech recognition result is not reliable, traditional structural modeling algorithms only using transcribed text do not perform well due to recognition errors by speech recognition system. To handle this challenge, we investigate sorts of features extracted from speech signal and transcribed documents and select the best ones to extract the hidden structure information from speech documents. We then propose novel algorithms based on traditional machine learning algorithms for structure extraction process and further using structure information to improve the performance of speech understanding and summariztion task. We will evaluate the proposed algorithms on the lecture conference speech corpus. On one hand, using our proposed algorithms can extract structure information hidden in speech documents more effectiv
英文关键词: structure modeling;speech document structure;speech understanding;speech summarization;extractive summarization