-
公开(公告)号：KR1020120067124A
公开(公告)日：2012-06-25
申请号：KR1020100128569
申请日：2010-12-15
Applicant: 한국전자통신연구원
Inventor： 이성주
IPC: G10L21/02
CPC classification number: G10L21/0272 , G10L21/0216
Abstract: PURPOSE: A multi-channel sound quality enhancement apparatus using similar auditory information and a running method thereof are provided to utilize soft-masking filter method using location information or direction of target sound source in noise environment. CONSTITUTION: A running method of multi-channel sound quality enhancement apparatus using similar auditory information comprises the following steps: presuming formal cross-correlation function[NCC] based on sound signal by multichannel NCC(Normalized Cross-Correlation) estimation unit(130); identifying one or more of signal among interference signal or background noise signal from the sound signal by a signal decision unit(135); presuming NCC density distribution of identified interference signal or background noise signal by a density estimation unit(140); and presuming soft-mask information by a soft-mask estimation unit(145).
Abstract translation: 目的：提供使用类似听觉信息的多声道音质增强装置及其运行方法，以利用噪声环境中使用位置信息或目标声源方向的软屏蔽滤波方法。 构成：使用类似听觉信息的多声道音质增强装置的运行方法包括以下步骤：基于多通道NCC（归一化交叉相关）估计单元（130）的声音信号，假定形式互相关函数[NCC]。 通过信号判定单元（135）识别来自声音信号的干扰信号或背景噪声信号中的一个或多个信号; 假设由密度估计单元（140）识别的干扰信号或背景噪声信号的NCC密度分布; 并且通过软掩模估计单元（145）假设软掩模信息。
-
公开(公告)号：KR1020120066523A
公开(公告)日：2012-06-22
申请号：KR1020100127898
申请日：2010-12-14
Applicant: 한국전자통신연구원
IPC: G10L15/18
Abstract: PURPOSE: A voice recognition system for personal customized natural language is provided to create various voice searching services through vocalization of the natural language. CONSTITUTION: A voice recognition system comprises: a control unit(123) which provides a customized model to a voice recognition unit(143) in case that a user is registered and controls provision of the customized model in cast that the user is not registered; and a service processing unit(133) which controls updating locutionary act and voice recognition result in case that the user agrees the result.
Abstract translation: 目的：提供个人定制自然语言的语音识别系统，通过自然语言的发声来创建各种语音搜索服务。 构成：语音识别系统包括：控制单元，其在用户注册的情况下向语音识别单元（143）提供定制模型，并控制用户未注册的定制模型的提供; 以及在用户同意结果的情况下控制更新定位动作和语音识别结果的服务处理单元（133）。
-
公开(公告)号：KR101082840B1
公开(公告)日：2011-11-11
申请号：KR1020080125433
申请日：2008-12-10
Applicant: 한국전자통신연구원
IPC: G10L99/00
Abstract: 본발명은음성과잡음신호분리방법및 그장치에관한것으로, 음원의통계적정보를이용하는음원분리기술과음원의공간적정보를활용하는빔포밍기술을두개이상의마이크로폰을갖춘시스템에사용할경우음성신호와잡음신호를보다효과적으로분리할수 있게되며, 결과적으로잡음환경에서녹음된신호로부터잡음신호가제거된깨끗한음성신호를추출할수 있다. 또한, 본발명은암묵신호분리기술에있어서학습과정이불필요하므로계산량이적고잘못된학습으로인한성능저하의염려가없는등, 음원분리의성능을높일뿐만아니라동시에가중치학습단계에서수렴속도를높임으로서계산효율성도제고할수 있으며, 빔포밍기술의경우에도일반적으로알려지지않은잡음원의개수및 위치에관계없이환경에강인하게동작할수 있다.
-
公开(公告)号：KR101072888B1
公开(公告)日：2011-10-18
申请号：KR1020080131239
申请日：2008-12-22
Applicant: 한국전자통신연구원
Abstract: 본발명은채널추정기반변별학습을이용한환경적응방법에관한것으로, 음성인식을다양한환경에적용할때 각환경으로의적응을위한효과적인방법을제공하며, 변별학습기반에서모델적응기법을수용하는방식으로일차적으로변별력을유지하는적응데이터에대해채널특성을찾아모델을변환을수행하고이를이용하여변별학습기법과결합하는방식으로효과적인환경적응을제공할수 있는이점이있다.
-
公开(公告)号：KR101068122B1
公开(公告)日：2011-09-28
申请号：KR1020080126924
申请日：2008-12-15
Applicant: 한국전자통신연구원
Abstract: 본 발명은 음성인식기에서 가비지 및 반단어 모델 기반의 거절 기술에 관한 것으로, 특히 비음성을 거절하기 위한 가비지 모델(garbage model), 음소 유사도에 기반하는 반단어 모델(anti-word model) 구성법, 이들을 통합한 거절 네트워크, 거절 네트워크에 대한 고속 재평가를 위한 인접 프레임 간의 유사도에 근거한 프레임 제거법(frame dropping)을 동원하여 인식된 결과를 거절하는 것을 특징으로 한다. 본 발명에 의하면, 종래 음성인식을 위한 발성사전에 등록되어 있지 않은 미등록 어휘나 비문법적 어휘의 입력뿐만 아니라, 등록되지 않은 음향-음성학적 입력 신호의 입력에 대해 효과적인 거절 기능을 수행할 수 있으며 고속의 거절평가가 가능해짐으로써 인식성공률이나 반응시간에서 음성인식기의 성능 향상을 도모할 수 있다.
음성인식, 거절(rejection), 프레임 제거법, 가비지 모델, 반단어 모델-
Patent Agency Ranking
公开(公告)号：KR1020110071742A
公开(公告)日：2011-06-29
申请号：KR1020090128386
申请日：2009-12-21
Applicant: 한국전자통신연구원
Abstract: PURPOSE: An utterance verification apparatus based on a word reliability threshold and a method thereof are provided to apply different reliability threshold to each word recognized in a word-based utterance verification system with respect to a voice recognition result. CONSTITUTION: A phoneme segment information extractor(130) extracts phoneme segment information with the analysis of a recognized word. Likelihood value calculators(140,150) calculate an likelihood value for the extracted phoneme and half-phoneme. A threshold calculator(170) calculates a threshold value corresponding to the recognized word. A comparator(190) compares the threshold value with an LLR(Log Likelihood Ratio) calculated by the likelihood value calculator. According to a comparison result, the comparator outputs or secludes a voice recognition result.
Abstract translation: 目的：提供一种基于字可靠性阈值的话语验证装置及其方法，以针对语音识别结果对基于词语的话语验证系统中识别的每个单词应用不同的可靠性阈值。 构成：音素段信息提取器（130）通过识别字词的分析来提取音素段信息。 似然值计算器（140,150）计算提取的音素和半音素的似然值。 阈值计算器（170）计算与所识别的字对应的阈值。 比较器（190）将阈值与由似然值计算器计算的LLR（对数似然比）进行比较。 根据比较结果，比较器输出或隐藏语音识别结果。
-
公开(公告)号：KR1020110071738A
公开(公告)日：2011-06-29
申请号：KR1020090128382
申请日：2009-12-21
Applicant: 한국전자통신연구원
Abstract: PURPOSE: A material-distribution search method using a voice recognizing function and a method thereof are provided to search a proper transportation request by searching a distribution in the moving route in real time. CONSTITUTION: The location of a truck is traced and a truck owner selects a freight search mode through a voice(202). If the selected freight search mode is an automatic search mode, the truck owner selects the forward data search or the backward data search through a voice(203,204). If the forward data search is selected, the transportation request information after the current location is provided(205). If the backward data search is selected, the transportation request information prior to the current location is provided(206).
Abstract translation: 目的：提供一种使用语音识别功能的材料分配搜索方法及其方法，用于通过实时搜索移动路线中的分布来搜索适当的运输请求。 宪法：追查卡车的位置，货车所有人通过声音选择货运搜索模式（202）。 如果所选择的货运搜索模式是自动搜索模式，则卡车所有者通过语音选择前向数据搜索或反向数据搜索（203,204）。 如果选择正向数据搜索，则提供当前位置之后的运送请求信息（205）。 如果选择了反向数据搜索，则提供当前位置之前的运输请求信息（206）。
-
公开(公告)号：KR1020110071297A
公开(公告)日：2011-06-29
申请号：KR1020090127819
申请日：2009-12-21
Applicant: 한국전자통신연구원
IPC: G10L15/183 , G10L15/28
Abstract: PURPOSE: A voice recognition apparatus for generating a plurality of recognition results is provided to generate N-best recognition results using a phoneme column-based search unit. CONSTITUTION: A continuous voice recognition unit(101) performs the voice recognition of input voice data. The continuous voice recognition unit outputs a word column which is most similar to the input voice data as a recognition result. A phoneme column converter(102) changes the recognition result into a phoneme column. A phoneme column-based search unit(103) searches a plurality of word columns of which a phoneme column distance is contiguity with the recognition result from a language model(105).
Abstract translation: 目的：提供用于产生多个识别结果的语音识别装置，以使用基于音素列的搜索单元产生N最佳识别结果。 构成：连续语音识别单元（101）执行输入语音数据的语音识别。 连续语音识别单元输出与输入语音数据最相似的字列作为识别结果。 音素列转换器（102）将识别结果改变为音素列。 基于音素列的搜索单元（103）搜索与语言模型（105）的识别结果相邻的音素列距离的多个单词列。
-
19.发明公开
公开(公告)号：KR1020110070688A
公开(公告)日：2011-06-24
申请号：KR1020100033376
申请日：2010-04-12
Applicant: 한국전자통신연구원
Abstract: PURPOSE: A voice recognition apparatus and method having two-step utterance verification structure for reducing the complexity of N-best recognized word calculation are provided to induce the re-utterance of a user or notify the user of an utterance error. CONSTITUTION: Using a first model, a voice recognition module(130) recognizes the voice of input voice data. The voice recognition module outputs a first N-best word list. An utterance verification module(140) creates a second N-best word list. Using a second model, the utterance verification module creates a final N-best word list from the second N-best word list.
Abstract translation: 目的：提供具有用于降低最佳识别字计算的复杂度的两步话语验证结构的语音识别装置和方法，以引起用户的重新发音或通知用户话语错误。 构成：使用第一模型，语音识别模块（130）识别输入语音数据的语音。 语音识别模块输出第一个N最好的单词列表。 话音验证模块（140）创建第二N个最佳词列表。 使用第二个模型，话语验证模块从第二个N最佳词列表中创建一个最后的N个最佳词列表。
-
公开(公告)号：KR1020110038447A
公开(公告)日：2011-04-14
申请号：KR1020090095740
申请日：2009-10-08
Applicant: 한국전자통신연구원
Inventor： 이성주
CPC classification number: G10L25/78 , G10L15/14 , G10L15/20 , G10L21/0208 , G10L2021/02166
Abstract: PURPOSE: A target signal detecting device using a statistical model and a method thereof are provided to irrelevantly detect a voice frame interval where a voice of a user exists in a noise environment. CONSTITUTION: A cross correlation function estimation unit(23-1) calculates a conditional probabilities about a plurality of sound source frame corresponding to an audio signal. The cross correlation function estimating unit estimates a likelihood ratio of a conditional unit probability in case of absence and a case of a target signal existing about a cross correlation function which is normalized through the conditional unit probabilities. A density estimating unit(25) estimates density in moving average about the cross correlation function. A interference signal density estimation unit(29) estimates statistical average and deviation of the normalized cross correlation function having an interference signal frame in the conditional unit target signal absence probability.
Abstract translation: 目的：提供使用统计模型的目标信号检测装置及其方法，以便在噪声环境中存在用户的语音的语音帧间隔无关地检测。 构成：互相关函数估计单元（23-1）计算关于与音频信号对应的多个声源帧的条件概率。 互相关函数估计单元估计在缺席情况下的条件单位概率的似然比，以及关于通过条件单位概率归一化的互相关函数存在目标信号的情况。 密度估计单元（25）估计关于互相关函数的移动平均值的密度。 干扰信号密度估计单元（29）估计在条件单位目标信号不存在概率中具有干扰信号帧的归一化互相关函数的统计平均和偏差。
-
-
-
-
-
-
-
-
-
Search Results
127
records
(took 0.028 seconds)
