专利快速检索-快速检索全球专利，免费商用专利数据库-IPRDB

21. 发明公开

KR1020150001191A 연속어 음성 인식 장치 및 방법 审中-实审
标题翻译：用于识别连续语音的装置和方法
公开(公告)号：KR1020150001191A
公开(公告)日：2015-01-06
申请号：KR1020130073990
申请日：2013-06-26
申请人： 한국전자통신연구원
发明人： 박기영 , 정훈 , 이윤근
IPC分类号： G10L15/08 , G10L15/02
CPC分类号： G10L15/18 , G10L15/04 , G10L15/32
摘要： 본 발명은 대용량 어휘가 포함된 연속어에 대하여 음성 인식을 수행하는 장치 및 방법에 관한 것이다. 본 발명은 동일 종류의 어휘의 수가 매우 많은 대어휘 연속어에서 대어휘를 적당한 개수의 클러스터로 나누고, 해당 클러스터의 대표 어휘를 선정하여 대표 어휘만 포함된 1단계 인식을 수행하며, 1단계 수행한 결과를 이용하여 대표 어휘가 인식되었을 경우 인식된 대표 어휘가 속한 클러스터 내의 모든 단어에 대하여 재인식을 수행한다.
摘要翻译：本发明涉及一种用于在包括大词汇的连续语音上执行语音识别的装置和方法。在本发明中，在包含大量相同类型的词汇的大词汇连续语音中，大词汇被划分成适当数量的群集，确定相应簇的代表性词汇表，并且包括第只有代表性的词汇被执行。当通过使用执行第一步骤的结果来识别代表性词汇时，对识别的代表词汇所属的集群内的所有单词执行重新识别。

22. 发明公开

KR1020140139922A 음성 인식 장치 및 이의 음성 인식 방법 审中-实审
标题翻译：语音识别装置及其识别方法
公开(公告)号：KR1020140139922A
公开(公告)日：2014-12-08
申请号：KR1020130060637
申请日：2013-05-28
申请人： 삼성전자주식회사
发明人： 박호연
IPC分类号： G10L15/02 , G10L15/28
CPC分类号： G10L15/04 , G10L15/22 , G10L15/30 , G10L2015/221
摘要： 음성 인식 장치 및 이의 음성 인식 방법이 제공된다. 본 음성 인식 장치의 음성 인식 방법은 사용자가 발화한 음성에 대응되는 발화 데이터를 수신하고, 발화 데이터에 대해 음성 인식을 수행하여 복수의 인식 결과를 생성하며, 생성된 복수의 인식 결과 중 적어도 하나를 외부의 서버에 실시간으로 전송한다.
摘要翻译：提供语音识别装置及其语音识别方法。语音识别装置的语音识别方法包括以下步骤：接收对应于用户语音的语音数据; 通过对语音数据进行语音识别来产生多个识别结果; 以及将所述多个生成的识别结果中的至少一个实时地发送到外部服务器。

23. 发明公开

KR1020140076145A 잡음 환경에서의 음성 인식을 위한 특징 벡터 정규화 장치 및 방법 审中-实审
标题翻译：用于通知噪声环境中语音识别的特征向量的装置及其方法
公开(公告)号：KR1020140076145A
公开(公告)日：2014-06-20
申请号：KR1020120144356
申请日：2012-12-12
申请人： 한국전자통신연구원
发明人： 박기영 , 이윤근 , 박전규
IPC分类号： G10L15/20 , G10L19/00
CPC分类号： G10L19/038 , G10L15/04 , G10L15/20
摘要： The present invention relates to an apparatus to normalize feature vectors for speech recognition in a noisy environment and a method thereof. The present invention relates to a technique which secures rapid response speed in a silent environment by combining a total normalization method of high recognition performance with an online normalization method to increase the response speed when a feature vector normalization method is applied and can prevent the deterioration of the performance for speech recognition in a noisy environment.
摘要翻译：本发明涉及一种用于在噪声环境中对用于语音识别的特征向量进行归一化的装置及其方法。本发明涉及通过组合高识别性能的总规范化方法与在线规范化方法来确保无声环境中的快速响应速度的技术，以在应用特征向量归一化方法时提高响应速度，并且可以防止恶化在嘈杂的环境中表演语音识别。

24. 发明公开

KR1020140035164A 음성인식시스템의 동작방법 无效
标题翻译：语音识别系统的方法操作
公开(公告)号：KR1020140035164A
公开(公告)日：2014-03-21
申请号：KR1020120101690
申请日：2012-09-13
申请人： 한국전자통신연구원
发明人： 오유리 , 강점자 , 이윤근
IPC分类号： G10L15/06
CPC分类号： G10L15/06 , G10L15/04 , G10L15/08
摘要： One embodiment provides a method operating speech recognition system which includes a step of recognizing at least speech including speech information based on a prior sound model provided from an evaluation data base, a step of sorting correct recognition speeches and incorrect recognition speech among at least on speech, a step of generating the reliability set of the incorrect speeches or the reliability of reference correct speeches formerly obtained by measuring each reliability of the incorrect speeches and each reliability of the correct speeches based on reference incorrect speeches or the reference correct speeches formerly obtained, and a step of obtaining a speech verification parameter including at least one among the number of the incorrect speeches, the incorrect speech reliability, the number of the correct speeches, the correct speech reliability with regard to an arbitrary speech based on the reliability set of the incorrect speeches and the reliability set of the incorrect speech. [Reference numerals] (AA) Start; (BB) End; (S110) Recognizing at least one speech including speech information; (S120) Sorting correct recognition speeches and incorrect recognition speech among at least one speech; (S130) Generating the reliability set of the incorrect speeches or the reliability of reference correct speeche; (S140) Obtaining a speech verification parameter
摘要翻译：一个实施例提供了一种操作语音识别系统的方法，其包括基于从评估数据库提供的先前的声音模型至少识别包括语音信息的语音的步骤，至少在语音上排列正确的识别语音和不正确的识别语音的步骤通过基于参考不正确的演讲或以前获得的参考正确的演讲，测量以前通过测量不正确的演讲的每个可靠性和正确演讲的每个可靠性而获得的不正确的演讲的可靠性集合或参考正确讲话的可靠性的步骤，以及基于不正确的可靠性集，获得语音验证参数的步骤，该语音验证参数包括不正确的演讲的数量，不正确的语音可靠性，正确的演讲的数量，关于任意语音的正确的语音可靠性演讲和可靠性设置的inco 直言不讳（附图标记）（AA）开始; （BB）结束; （S110）识别包括语音信息的至少一个语音; （S120）在至少一个演讲中排序正确的识别演讲和不正确的识别语音; （S130）生成不正确演讲的可靠性集或参考正确讲话的可靠性; （S140）获取语音验证参数

25. 发明公开

KR1020130134620A 디코딩 정보를 이용한 끝점 검출 장치 및 그 방법 无效
标题翻译：用于使用解码信息检测端点的装置和方法
公开(公告)号：KR1020130134620A
公开(公告)日：2013-12-10
申请号：KR1020120058249
申请日：2012-05-31
申请人： 한국전자통신연구원
发明人： 정훈 , 박기영 , 이성주 , 이윤근
IPC分类号： G10L25/93
CPC分类号： G10L15/05 , G10L15/04
摘要： The present invention provides a device for detecting an end point using decoding information comprising; an end point detector which detects the end point of an audio signal included in a sound signal by receiving the audio signal from the outside; a decoder which produces decoding information by decoding the audio signal; an end point discriminator which extracts reference information which becomes a standard of the determination of an actual end point and determines whether the end point detected in the end point detector is the actual end point among detected end points or not based on the extracted standard information. [Reference numerals] (110) End point detector;(120) Decoder;(130) End point discriminator;(AA) Audio signal input
摘要翻译：本发明提供一种使用解码信息检测终点的装置，包括：端点检测器，其通过从外部接收音频信号来检测包括在声音信号中的音频信号的终点; 解码器，通过解码音频信号产生解码信息; 提取作为实际终点的确定的标准的参照信息的终点鉴别器，根据所提取出的标准信息，判定检测结束点中检测到的终点是否为检测结束点之间的实际终点。（110）端点检测器;（120）解码器;（130）终点识别器;（AA）音频信号输入

26. 发明公开

KR1020130088637A 디스플레이장치 및 그 음성인식방법 审中-实审
标题翻译：显示装置及其语音识别方法
公开(公告)号：KR1020130088637A
公开(公告)日：2013-08-08
申请号：KR1020120010008
申请日：2012-01-31
申请人： 삼성전자주식회사
发明人： 김정근 , 고창석 , 최준식
IPC分类号： G10L15/00 , G10L15/26
CPC分类号： G10L15/22 , G10L15/04 , G10L15/26 , G10L15/28
摘要： PURPOSE: A display apparatus and a voice recognition method thereof are provided to perform a control operation corresponding to the selected voice information, by acquiring first and second voice information corresponding to a first and a second voice signal respectively, and adding a fixed weighted value to a voice signal with larger measured sound pressure, and then selecting one appropriate for the utterance intend of the user of the first and the second voice information. CONSTITUTION: A first voice acquisition unit (110) converts voice of a user into a first electrical voice signal. A second voice acquisition unit (120) converts the voice of the user into a second electrical voice signal together with the first voice acquisition unit. A sound pressure measurement unit (130) measures the sound pressure amplitude of the first and the second voice signal. A controller (140) acquires first and second voice information corresponding to the first and the second voice signal respectively, and selects one for the utterance intend of the user of the first and the second voice information. [Reference numerals] (110) First voice acquisition unit; (120) Second voice acquisition unit; (130) Sound pressure measurement unit; (140) Controller; (141) Voice recognition unit; (143) Determination unit; (145) Voice-letter conversion unit; (150) User input unit; (160) Display unit
摘要翻译：目的：提供一种显示装置及其语音识别方法，通过分别获取对应于第一和第二语音信号的第一和第二语音信息，并将固定加权值加到具有较大测量声压的语音信号，然后选择适合于第一和第二语音信息的用户的话语的一个。构成：第一语音获取单元（110）将用户的语音转换为第一电语音信号。第二语音获取单元（120）将用户的语音与第一语音获取单元一起转换为第二电语音信号。声压测量单元（130）测量第一和第二声音信号的声压幅度。控制器（140）分别获取与第一和第二语音信号相对应的第一和第二声音信息，并且为第一和第二语音信息的用户的话语选择一个。（附图标记）（110）第一语音获取单元; （120）第二语音获取单元; （130）声压测量单元; （140）控制器; （141）语音识别单元; （143）确定单位; （145）语音信号转换单元; （150）用户输入单元; （160）显示单元

27. 发明公开

KR1020130068622A 엔베스트 정보 기반 문맥 종속 반음소 모델을 이용한 발화검증 장치 및 방법 无效
标题翻译：使用基于N-BEST信息的上下文相对抗电文模型验证UTTERANCE的装置和方法
公开(公告)号：KR1020130068622A
公开(公告)日：2013-06-26
申请号：KR1020110135917
申请日：2011-12-15
申请人： 한국전자통신연구원
发明人： 강점자 , 정호영 , 이윤근
IPC分类号： G10L15/14 , G10L15/28
CPC分类号： G10L15/187 , G10L15/02 , G10L15/04 , G10L15/14
摘要： PURPOSE: An utterance verification device and a method thereof using N-best information based context subordination anti-phoneme model are provided to improve utterance verification performance by performing utterance verification using a context dependent anti-phoneme model generated by an N-best recognition result. CONSTITUTION: A pre-processing unit(140) extracts an end point and a feature from an input voice. A voice recognition unit(150) recognizes the voice based on context dependent phoneme model in a storage unit(120). An utterance verification unit(160) measures reliability based on the voice recognition result of the voice recognition unit and the context dependent phoneme model and context dependent anti-phoneme model. [Reference numerals] (110) Generation unit; (120) Saving unit; (130) Input unit; (140) Pre-processing unit; (150) Voice recognition unit; (160) Ignition verification unit
摘要翻译：目的：提供一种使用基于N最佳信息的上下文从属反音素模型的话语验证装置及其方法，以通过使用由N最佳识别结果生成的上下文相关反音素模型进行话音验证来改善话语验证性能。构成：预处理单元（140）从输入语音提取终点和特征。语音识别单元（150）在存储单元（120）中基于上下文相关音素模型识别语音。话音验证单元（160）基于语音识别单元的语音识别结果和上下文相关音素模型和上下文相关的反音素模型来测量可靠性。（附图标记）（110）生成单元; （120）保存单位; （130）输入单元; （140）预处理单元; （150）语音识别单元; （160）点火验证单元

28. 发明授权

KR100829583B1 전화 통화 내용을 자동으로 메모하는 방법 및 장치, 통화목록 표시 방법 및 장치 有权
标题翻译：用于自动呼叫内容的方法和装置，以及显示呼叫列表的方法和装置
公开(公告)号：KR100829583B1
公开(公告)日：2008-05-14
申请号：KR1020060124107
申请日：2006-12-07
申请人： 삼성전자주식회사
发明人： 정소영 , 고상철 , 조정미 , 김정수
IPC分类号： H04M1/65 , G10L15/26 , G10L15/04
CPC分类号： H04M1/6505 , G10L15/04 , G10L15/26
摘要： A method and device for automatically noting down telephone call contents, a call list display method, and a device therefor are provided to display stored memo information for a user in case a call list is inquired about after a call is terminated, thus the user can confirm call contents together with call details through displayed core word text memos. Call zones for interpreting a voice signal are set(102-110). The voice signal transceived in the set call zones is interpreted to extract core words included in the transceived voice signal(112). The extracted core words are set to memos, so that the set memos are stored(114). When a call is terminated, the stored memos are displayed(116,118). It is determined whether the displayed memos are tagged(120). The displayed memos are modified by a user(122).
摘要翻译：提供一种用于自动注销电话呼叫内容的方法和设备，呼叫列表显示方法及其设备，用于在呼叫终止后询问呼叫列表的情况下向用户显示所存储的备忘录信息，从而用户可以通过显示的核心词文本备忘录确认通话内容和通话详细信息。呼叫区域用于解释语音信号（102-110）。在设定呼叫区域中收发的语音信号被解释为提取包含在收发的语音信号（112）中的核心词。提取的核心字被设置为备忘录，使得设置的备忘录被存储（114）。当呼叫终止时，显示存储的备忘录（116,118）。确定所显示的备忘录是否被标记（120）。所显示的备忘录被用户修改（122）。

29. 发明公开

KR1020080012491A 음성 구간 검출 방법 및 장치, 및 이를 이용한 음성 인식시스템 有权
标题翻译：使用动态编程的语音/静音间隔识别的方法和装置及其语音识别系统
公开(公告)号：KR1020080012491A
公开(公告)日：2008-02-12
申请号：KR1020060073386
申请日：2006-08-03
申请人： 삼성전자주식회사 , 노바우리스 테크놀러지스 리미티드
发明人： 장길진 , 김정수 , 브리들존에스 , 헌트멜빈제이
IPC分类号： G10L15/20 , G10L15/08 , G10L25/78 , G10L15/04
CPC分类号： G10L25/78 , G10L15/04
摘要： A method and an apparatus for detecting a voice section, and a voice recognition system using the same are provided not to recognize a spurious phone-sequence as user's actual voice, thereby detecting effectively the voice section. A method for detecting a voice section comprises the following steps of: updating section values, an upper level track indicating an energy level voice section and a lower level track indicating a noise section, for determining a voice and a noise to the energy level of a current frame(400); dividing the energy level of the current frame into energy levels based on the upper and lower level tracks(402); determining local voice sections based on the energy level divided by DP(Dynamic Programming)(404); and combining the local voice section with the entire voice section of an input signal(406).
摘要翻译：提供了一种用于检测语音部分的方法和装置以及使用该方法和语音识别系统的语音识别系统，其不能将伪电话序列识别为用户的实际语音，从而有效地检测语音部分。用于检测语音部分的方法包括以下步骤：更新部分值，指示能量级语音部分的上级轨道和指示噪声区段的下级轨道，用于将语音和噪声确定为当前帧（400）; 将当前帧的能级划分为基于上级和下级轨道（402）的能级; 基于由DP（动态编程）划分的能量水平（404）来确定本地语音段; 以及将本地语音部分与输入信号的整个语音部分组合（406）。

30. 发明公开

KR1020070091789A 음성 신호에서 음소를 인식하는 방법 및 그 시스템 有权
标题翻译：方位信号识别方法及其系统
公开(公告)号：KR1020070091789A
公开(公告)日：2007-09-12
申请号：KR1020060021352
申请日：2006-03-07
申请人： 삼성전자주식회사
发明人： 김현수
IPC分类号： G10L15/04
CPC分类号： G10L15/04 , G10L2015/025
摘要： A method for recognizing a phoneme in a voice signal and a system thereof are provided to calculate a probability distance based on the peak average and standard deviation of each frame for each divided phoneme and recognize a phoneme having the minimum probability distance as a corresponding phoneme, thereby being applied to various systems with a low computational load. A method for recognizing a phoneme in a voice signal comprises the following steps of: determining a boundary between respective phonemes in an inputted voice signal and segmenting the phonemes(200,205); determining a length of a frame for calculating peak statistics in the segmented phoneme and producing the peak statistics in the determined frame(220,225); calculating a probability distance from the peak statistics of the determined frame by referring to the peak statistics within a phoneme table which indicates standard peak statistics for each phoneme(230); and recognizing the phoneme of the phoneme table having the minimum probability distance as a corresponding phoneme(235).
摘要翻译：提供了一种用于识别语音信号中的音素及其系统的方法，用于根据每个划分的音素的每个帧的峰值平均值和标准偏差来计算概率距离，并将具有最小概率距离的音素识别为相应的音素，从而应用于具有低计算负荷的各种系统。用于识别语音信号中的音素的方法包括以下步骤：确定所输入语音信号中的各个音素之间的边界并分割音素（200,205）; 确定用于计算分段音素中的峰值统计的帧的长度，并在确定的帧中产生峰值统计（220,225）; 通过参考指示每个音素（230）的标准峰值统计的音素表内的峰值统计来计算来自确定帧的峰值统计量的概率距离; 并将具有最小概率距离的音素表的音素识别为相应的音素（235）。

你已经成功收藏专利！

检索式保存成功!

IPRDB

热门服务

关于我们

友情链接

联系方式