首页> 外国专利> KOREAN CONTINUOUS VOICE RECOGNITION DEVICE AND METHOD USING DEPENDENT GRAMMAR AS BACKWARD LANGUAGE MODEL AND AUTOMATIC TRANSLATING SYSTEM USING IT

KOREAN CONTINUOUS VOICE RECOGNITION DEVICE AND METHOD USING DEPENDENT GRAMMAR AS BACKWARD LANGUAGE MODEL AND AUTOMATIC TRANSLATING SYSTEM USING IT

机译:韩国语连续语音识别装置及方法,采用依存语法为后向语言模型,并采用其自动翻译系统

摘要

The present invention relates to an apparatus and method for recognizing Korean continuous speech and an automatic interpretation system using the same, using a conventional simple grammar as a forward search method and adopting a dependent syntax that expresses Korean grammar features well with a backward search method. Feature extraction means (11) for extracting features of an input speech to provide a speech recognition apparatus and method and an automatic interpretation system using the same; Word modeling means 15 for modeling words; Word recognition means (12) for recognizing speech on a word-by-word basis; Sentence recognizing means (13) for recognizing a sentence using a bigram 18 as a forward-looking language model; And sentence recognizing means (14) using the dependent grammar (19) as a backward language model, comprising: a first step (31, 32) of determining whether it is the last frame after initialization; A second step (33, 34) of performing a language processing process by applying a forward language model after performing Viterbi calculation on the input speech every frame unit; and backward language model every word unit from the last frame. A continuous speech recognition means (41, 42) comprising a third step (35) of applying the grammar and using the dependent grammar as a backward language model; Storage means 43 for storing parsing results analyzed by the backward language model; Conversion and generating means 44 for converting and generating a result of parsing into a sentence of another language; And a speech synthesis means 45 for synthesizing and outputting the translated speech, thereby improving continuous speech recognition rate, and having an economic effect by reducing interpretation time and simplifying the system.
机译:本发明涉及一种用于识别韩语连续语音的装置和方法以及使用该装置和方法的自动解释系统,该装置和方法使用常规的简单语法作为正向搜索方法,并采用一种通过向后搜索方法很好地表达韩语语法特征的从属语法。特征提取装置(11),用于提取输入语音的特征以提供语音识别装置和方法以及使用该装置的自动解释系统;单词建模装置15,用于对单词建模;单词识别装置(12),用于逐个单词地识别语音;句子识别装置(13),用于使用双字组18作为前瞻性语言模型来识别句子;以及使用从属语法(19)作为后向语言模型的句子识别装置(14),包括:确定是否是初始化之后的最后一帧的第一步(31、32);以及第二步(33、34),在对输入语音的每一帧单位进行维特比计算之后,通过应用前向语言模型来执行语言处理过程;后向语言则对最后一帧中的每个单词单元进行建模。连续语音识别装置(41、42),包括应用语法并将从属语法用作后向语言模型的第三步骤(35);存储装置43,用于存储由后向语言模型分析的解析结果;转换和生成装置44,用于将解析结果转换和生成为另一种语言的句子;语音合成装置45用于合成和输出翻译后的语音,从而提高连续语音识别率,并通过减少解释时间和简化系统而具有经济效果。

著录项

  • 公开/公告号KR0136425B1

    专利类型

  • 公开/公告日1998-05-15

    原文格式PDF

  • 申请/专利权人 KOREA TELECOM CORP.;

    申请/专利号KR19950001402

  • 发明设计人 구명완;장두성;

    申请日1995-01-26

  • 分类号G10L5/04;

  • 国家 KR

  • 入库时间 2022-08-22 02:47:28

相似文献

  • 专利
  • 外文文献
  • 中文文献
获取专利

客服邮箱:kefu@zhangqiaokeyan.com

京公网安备:11010802029741号 ICP备案号:京ICP备15016152号-6 六维联合信息科技 (北京) 有限公司©版权所有
  • 客服微信

  • 服务号