-
公开(公告)号：KR1020130013128A
公开(公告)日：2013-02-06
申请号：KR1020110074552
申请日：2011-07-27
Applicant: 포항공과대학교 산학협력단
CPC classification number: G09B17/02 , G09B5/02 , G09B19/04 , G10L13/086
Abstract: PURPOSE: A corpus generator for practicing phrase break prediction, and a method thereof are provided to minimize the reflection of each transcriber`s characteristic, thereby maximizing the reflection of the general transcribers` characteristics of the predicted phrase break. CONSTITUTION: A corpus generator(100) for practicing phrase break prediction comprises: A boundary extractor(110), a boundary determiner(120), and a phrase break prediction corpus generator(130). The boundary extractor receives multiple transcription data transcribed after phrase boundaries about the same text from each of the multiple labeled transcribers and extracts phrase boundaries from each of the multiple transcription data. The boundary determiner compares each of the extracted phrase boundaries, and determines phrase boundaries satisfied with a fixed setting to a phrase break prediction. The phrase break prediction corpus generator labels the determined phrase break prediction, and generates the phrase break prediction corpus. [Reference numerals] (1000) Phrase break prediction corpus DB; (110) Boundary extractor; (120) Boundary determiner; (121) First determination module; (122) Second determination module; (130) Phrase break prediction corpus generator; (AA) Transfer data
Abstract translation: 目的：提供一种用于练习短语中断预测的语料库生成器及其方法，以最小化每个转录器特征的反射，从而最大化预测短语中断的一般誊写者特征的反映。 构成：用于实施短语中断预测的语料库生成器（100）包括：边界提取器（110），边界确定器（120）和短语中断预测语料库生成器（130）。 边界提取器从多个标记的转录器中的每一个接收关于相同文本的短语边界之后转录的多个转录数据，并从多个转录数据中的每一个提取短语边界。 边界确定器比较每个提取的短语边界，并将满足固定设置的短语边界确定为短语中断预测。 短语break预测语料库生成器标记确定的短语中断预测，并生成短语break预测语料库。 [参考数字]（1000）短语突破预测语料库DB; （110）边界提取器; （120）边界确定因子 （121）第一确定模块; （122）第二确定模块; （130）短语预测语料库发生器; （AA）传输数据
-
公开(公告)号：KR1020140068760A
公开(公告)日：2014-06-09
申请号：KR1020130142742
申请日：2013-11-22
Applicant: 포항공과대학교 산학협력단
IPC: G06F17/00
Abstract: Disclosed are a dialog management method using memory ability and a device thereof. The dialog management method using the memory ability according to the present invention comprises: a step of receiving an articulation sentence; a step of obtaining a first response by analyzing the intention of the articulation sentence; a step of obtaining a second response by analyzing the unit knowledge of the articulation sentence; and a step of outputting the final response based on the first response and the second response. The present invention is provided to build a dialog management database which reflects the interest and the concern of a conversion partner and to output a proper response for articulation inputted through the database.
Abstract translation: 公开了一种使用记忆能力的对话管理方法及其装置。 根据本发明的使用存储能力的对话管理方法包括：接收发音语句的步骤; 通过分析发音句的意图来获得第一反应的步骤; 通过分析关节语句的单位知识来获得第二响应的步骤; 以及基于第一响应和第二响应输出最终响应的步骤。 提供本发明以构建一个反映转换伙伴的兴趣和关注的对话管理数据库，并输出通过数据库输入的发音的适当响应。
-
公开(公告)号：KR1020130043817A
公开(公告)日：2013-05-02
申请号：KR1020110107940
申请日：2011-10-21
Applicant: 포항공과대학교 산학협력단
Abstract: PURPOSE: An apparatus for learning language and a method thereof are provided to offer feedback for voice input to a user by analyzing the voice input. CONSTITUTION: A first voice recognition part(110) generates a first transcribed material transcribing voice input by recognizing the voice input in a word unit. A second voice recognition part(120) generates a second transcribed material transcribing the voice input by recognizing the voice input in a phoneme unit. A standard pronunciation generation part(130) generates a third transcribed material by converting the first transcribed material into predefined standard pronunciation. An error detection part(140) generates error information for the part of the second transcribed material not satisfying a predetermined reference by comparing the second transcribed material with the third transcribed material. A feedback generation part(150) generates feedback information corresponding to the error information. [Reference numerals] (1000) Pronunciation data DB; (110) First voice recognition part; (120) Second voice recognition part; (130) Standard pronunciation generation part; (140) Error detection part; (150) Feedback generation part; (160) Expansion pronunciation generation part; (2000) Pronunciation dictionary DB; (3000) Feedback DB; (AA) Voice input
Abstract translation: 目的：提供一种用于学习语言的装置及其方法，以通过分析语音输入来提供对用户的语音输入的反馈。 构成：第一语音识别部分（110）通过识别单词中的语音输入来生成转录语音输入的第一转录素材。 第二语音识别部件（120）通过识别音素单元中的语音输入来生成转录语音输入的第二转录材料。 标准发音生成部件（130）通过将第一转录材料转换为预定义的标准发音来生成第三转录材料。 错误检测部件（140）通过将第二转录材料与第三转录材料进行比较来生成不满足预定参考的第二转录材料的部分的误差信息。 反馈生成部（150）生成与错误信息对应的反馈信息。 （附图标记）（1000）发音数据DB; （110）第一语音识别部分; （120）第二语音识别部分; （130）标准发音生成部分; （140）错误检测部分; （150）反馈生成部分; （160）扩音发音部分; （2000）发音词典DB; （3000）反馈DB; （AA）语音输入
-
Patent Agency Ranking
公开(公告)号：KR101672484B1
公开(公告)日：2016-11-03
申请号：KR1020150103092
申请日：2015-07-21
Applicant: 포항공과대학교 산학협력단
Abstract: 발음오류검출장치는, 입력음성을인식하여단어인식결과를생성하는단어인식부; 상기단어인식결과에대응하는음소인식정보를생성하는음소인식부; 상기음소인식정보를기초로상기입력음성의발음점수를계산하는발음점수계산부; 상기음소인식정보를이용하여상기입력음성의발음오류확률을계산하는오류확률계산부; 및상기단어인식결과, 상기발음점수및 상기발음오류확률을기초로상기입력음성에발음오류가있는지판단하는오류판단부를포함하고, 상기음소인식부는표준발음사전및 다중발음사전을참조하여상기음소인식정보를생성한다.
-
公开(公告)号：KR101565658B1
公开(公告)日：2015-11-04
申请号：KR1020130142742
申请日：2013-11-22
Applicant: 포항공과대학교 산학협력단
IPC: G06F17/00
Abstract: 기억능력을이용한대화관리방법및 장치가개시된다. 본발명에따른기억능력을이용한대화관리방법은발화문장을수신하는단계, 수신된발화문장의의도를분석하여제1 응답을획득하는단계, 수신된발화문장의단위지식을분석하여제2 응답을획득하는단계및 상기제1 응답및 제2 응답에기초하여최종응답을출력하는단계를포함한다. 본발명에의하면대화상대의흥미및 관심사를반영하는대화관리데이터베이스를구축하고, 상기데이터베이스를통하여입력되는발화에대한적합한응답을출력가능하다.
-
公开(公告)号：KR1020140054771A
公开(公告)日：2014-05-09
申请号：KR1020120120674
申请日：2012-10-29
Applicant: 포항공과대학교 산학협력단
IPC: G10L15/22
CPC classification number: G10L15/222 , G10L15/01
Abstract: A system and a method for improving an utterance recognition performance are disclosed. Specifically, a system for improving an utterance recognition performance comprises: a weight adjusting unit for applying a weight based on a training corpus which is a voice recognition language model with respect to an existing user utterance sentence and a current user utterance sentence inputted in language; a recognition N-BEST generating unit for generating an N-BEST and an N-BEST score with respect to each of the weighted existing user utterance sentence and the weighted current user utterance sentence; and a recognition N-BEST combining unit for adding up the scores of the existing user utterance sentence and the current user utterance sentence with respect to the N-BEST. Therefore, the user utterance sentence can be corrected conveniently and easily because the user need not repeat re-utterance so as to correct an inaccurate existing user utterance sentence.
Abstract translation: 公开了一种用于提高话语识别性能的系统和方法。 具体地说，一种用于改善话语识别性能的系统包括：权重调节单元，用于基于用语言输入的现有用户话语句和当前用户发话语言的语音识别语言模型的训练语料库应用权重; 识别N-BEST生成单元，用于生成相对于加权的现有用户发言语句和加权的当前用户发言语句中的每一个的N-BEST和N-BEST分数; 以及识别N-BEST组合单元，用于将相对于N-BEST的现有用户话语句和当前用户发声句的分数相加。 因此，用户话语句可以方便和方便地进行修正，因为用户不需要重复重复发音，从而纠正不准确的现有用户发声句。
-
公开(公告)号：KR101294882B1
公开(公告)日：2013-08-08
申请号：KR1020110097916
申请日：2011-09-28
Applicant: 포항공과대학교 산학협력단
Abstract: 본 발명의 목적은 하나의 입력에 대하여 복수의 답을 생성하기 위한 기계 학습 방법을 제공하는 것이다. 본 발명에 의하면, 변환규칙을 이용하여 하나의 입력으로부터 복수의 답을 생성할 수 있으므로, 기계 학습 방법 및 장치를 하나의 입력으로부터 복수의 답을 생성하는 문제에 적용할 수 있는 장점이 있다. 또한, 변환규칙에 의하여 생성되는 복수의 답 개수를 생성의 정확성에 따라 제한함으로써, 답의 과생성으로 인하여 계산이 복잡해지는 것을 방지할 수 있는 장점이 있다.
-
公开(公告)号：KR101261754B1
公开(公告)日：2013-05-07
申请号：KR1020110074552
申请日：2011-07-27
Applicant: 포항공과대학교 산학협력단
Abstract: 끊어읽기훈련용말뭉치생성장치및 방법이개시된다. 끊어읽기훈련용말뭉치생성장치는복수의전사자각각으로부터동일한텍스트에대한어절경계가레이블링되어전사된복수개의전사자료를제공받아복수개의전사자료각각에서어절경계를추출하는경계추출부와, 추출된어절경계각각을비교하여미리설정된기준에만족되는어절경계를텍스트의끊어읽기로결정하는경계결정부와, 결정된끊어읽기를레이블링하여끊어읽기훈련용말뭉치를생성하는끊어읽기말뭉치생성부를포함하여구성된다. 따라서, 끊어읽기훈련용말뭉치가동일한텍스트에대한복수의전사자로부터전사된복수개의전사자료를이용하여형성되므로개별전사자의특성의반영을최소한으로하여보편적인전사자의끊어읽기특성을최대한반영할수 있으므로평균적으로높은성능을나타낼수 있다.
-
-
-
-
-
-
-
Search Results
18
records
(took 0.02 seconds)
