语音关键词检测中置信测度方法研究综述  

An Overview of Confidence Measure for Spoken Term Detection

作　　者：李海洋[1] 韩纪庆[1] 郑贵滨[1] 郑铁然[1] 

机构地区：[1]哈尔滨工业大学计算机科学与技术学院,哈尔滨150001

出　　处：《智能计算机与应用》2014年第2期10-15,共6页Intelligent Computer and Applications

基　　金：国家重点基础研究发展计划(973)(2007CB311104);国家自然科学基金重大研究计划项目(91120303)

摘　　要：语音关键词检测是指在语音文档中寻找并定位特定的词的技术,输入所需查询的关键词通常是以文本的形式给出。作为语音文档分析等技术的核心部分,语音关键词检测始终是语音处理领域研究的热点。置信测度计算是关键词检测技术的重要组成部分,对确认正确检出的关键词及拒绝误识都起到决定性作用,置信测度的优劣对检测系统性能有着直接的影响。介绍并总结了语音关键词检测中测置信测度方法的研究工作,提供了详尽的参考文献。Spoken term detection（ STD） is the task which aims to locate all occurrences of terms queried by a user in large audio archives,and the terms are usually in the format of text. STD is one of the key components for spoken document analysis,and STD is always a focus in the field of speech processing. Confidence measure is a crucial part of STD,and it plays an important role in rejecting false alarms. The merit of confidence measure can affect the performance of STD directly. This paper introduces and summarizes the research work of confidence measure for STD,and provides with plenty of references.

关 键 词：检测 置信测度 词表外词 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]