-
公开(公告)号：KR101217524B1
公开(公告)日：2013-01-18
申请号：KR1020080131755
申请日：2008-12-22
Applicant: 한국전자통신연구원
CPC classification number: G10L15/187 , G10L15/10
Abstract: 본 발명은 고립어 엔베스트 인식결과를 위한 발화검증 방법 및 장치에 있어서, 엔베스트 음성인식을 통해 단어수준에서 인식된 결과에 대해 엔베스트 발화검증을 통해서 신뢰도를 측정하고, 동적 시간 와핑에 의한 음소간 유사도를 측정하여 신뢰도와 유사도를 기반으로 음성인식의 수락/거절 또는 판단불가 등을 표시함으로써 보다 신뢰성 높은 음성인식이 가능하도록 한다.
엔베스트, 발화검증, 음성, 인식, 신뢰도, 유사도-
公开(公告)号：KR101095864B1
公开(公告)日：2011-12-21
申请号：KR1020090027532
申请日：2009-03-31
Applicant: 한국전자통신연구원
Abstract: 본 발명은 연속으로 발성되는 한국어 숫자음성에 대한 음성인식을 수행하여 그 인식 결과에 대해 혼동행렬과 신뢰도 치에 기반하여 다수의 음성인식 후보를 생성하는 기술에 관한 것으로, 혼동행렬은 숫자음성에 대해 오인식이 발생하는 숫자들로 구성되므로, 사전에 실험용 데이터베이스를 사용하여 인식을 수행한다. 또한 음성인식의 결과로 도출되는 숫자별 인식 점수인 통계적 우도를 그 단어의 지속시간 지수인 프레임 수로 나눈 로그 우도비를 신뢰도 치로써 사용하는 것을 특징으로 한다. 본 발명에 의하면, 음성인식 알고리즘에서 성능의 저하 없이 N-best를 생성하기 위해 사용하는 기억장치 사용량과 탐색시간을 절약함으로써 음성인식 엔진의 효율을 높일 수 있다.
음성인식, 숫자음성, N-best, 신뢰도치, 혼동행렬-
公开(公告)号：KR101073934B1
公开(公告)日：2011-10-17
申请号：KR1020080131759
申请日：2008-12-22
Applicant: 한국전자통신연구원
Abstract: 본발명에서는음성과음악이혼합된오디오신호로부터음성과음악을판별함에있어서, 입력오디오신호의프레임으로부터 MFCC(Mel-Frequency Cepstral Coefficient) 특징벡터와음성과음악의판별을위한새로운특징벡터값을추출한후, 각각의특징벡터별로구성되는 GMM 모델을이용한유사도를검사하고, 유사도검사값을이용하여 2차의복합특징벡터값을구성한후, 이를이용하고해당 GMM과비교하는다단계검증과정을거쳐입력오디오신호의구간별음성과음악을판별한다.
-
公开(公告)号：KR1020110070496A
公开(公告)日：2011-06-24
申请号：KR1020090127336
申请日：2009-12-18
Applicant: 한국전자통신연구원
CPC classification number: G06F17/30017 , G06F3/041 , G06F9/542 , G06F17/30244 , G06F17/3087 , G06Q50/10
Abstract: PURPOSE: Interactive contents providing device and method in an e-book system are provided to interactive contents by editing the contents provided from CP. CONSTITUTION: CP(Contents Provider)(102,103) generates and provides the contents. A contents making device(110) creates interactive contents by manufacturing and editing the contents received from the CP. The contents making device provides to the interactive contents to the CP. Terminals(105-108,119) receive the interactive contents from the CP and provide the interactive contents to users. The interactive contents comprise script, object data and scene data.
Abstract translation: 目的：通过编辑CP提供的内容，将电子书系统中的互动内容提供设备和方法提供给交互式内容。 构成：CP（内容提供者）（102,103）生成并提供内容。 内容制作装置（110）通过制造和编辑从CP接收的内容来创建交互式内容。 内容制作设备提供给CP的交互内容。 终端（105-108,119）从CP接收交互内容，并向用户提供交互式内容。 交互内容包括脚本，对象数据和场景数据。
-
公开(公告)号：KR1020110057525A
公开(公告)日：2011-06-01
申请号：KR1020090113966
申请日：2009-11-24
Applicant: 한국전자통신연구원
IPC: G10L21/0272 , G10L15/20
Abstract: PURPOSE: A device for separating a sound source and a method thereof are provided to extract only a desired sound from various sound sources. CONSTITUTION: An input unit(610) changes the offered signal in to a frequency domain. A processing unit(620) divides the sound source of the converted signal in the frequency band unit. The processing unit aligns the separated sound source through the phase difference of a mixed filter for mixing the sound sources. An output unit(630) changes the aligned sound sources into the time domain.
Abstract translation: 目的：提供一种用于分离声源的装置及其方法，以从各种声源提取期望的声音。 构成：输入单元（610）将提供的信号改变为频域。 处理单元（620）以频带单位划分转换信号的声源。 处理单元通过用于混合声源的混合滤波器的相位差来对准分离的声源。 输出单元（630）将对准的声源改变为时域。
-
Patent Agency Ranking
公开(公告)号：KR1020110035170A
公开(公告)日：2011-04-06
申请号：KR1020090092779
申请日：2009-09-30
Applicant: 한국전자통신연구원
IPC: G10L21/0208 , G10L15/20 , G10L15/14
CPC classification number: G10L21/0208 , G10L15/20 , G10L21/0216 , G10L21/0232 , G10L25/78
Abstract: PURPOSE: An apparatus for filtering a noise based on a model by compensating distortion to recognize a voice are provided to remove a noise without distortion for voice recognition. CONSTITUTION: A voice member probability calculator(206) calculates a voice absence probability, and a noise estimating and updating unit(208) updates the estimated noise. A first noise cancellation filter outputs a first pure voice included in the distortion through a filtering operation by using the voice absence probability and the updated estimated noise. A second noise removing filter outputs a distortion-compensated final voice signal by filtering pure voice estimation. The pure voice estimation value is obtained based on posteriori probability.
Abstract translation: 目的：提供一种用于通过补偿失真来识别语音来基于模型对噪声进行滤波的装置，以消除噪声，而不会使声音识别失真。 构成：语音成员概率计算器（206）计算语音缺失概率，并且噪声估计和更新单元（208）更新估计的噪声。 第一噪声消除滤波器通过使用语音缺失概率和更新的估计噪声通过滤波操作输出包括在失真中的第一纯音。 第二噪声去除滤波器通过对纯语音估计进行滤波来输出失真补偿的最终话音信号。 基于后验概率获得纯语音估计值。
-
公开(公告)号：KR1020100072838A
公开(公告)日：2010-07-01
申请号：KR1020080131365
申请日：2008-12-22
Applicant: 한국전자통신연구원
CPC classification number: G10L15/08 , G10L15/142
Abstract: PURPOSE: A viterbi decoder and a method for recognizing a voice are provided to prevent the dramatic lowering of an observation probability of a contaminated portion caused by an unintended impulse noise. CONSTITUTION: An optimal state calculator(220) obtains the state of the maximum accumulated similarity in each measurement vector of an observation vector row for the inputted voice. A buffer unit(240) stores an observation probability value for the plural voices inputted prior to the inputted voice. A non-linear filtering unit(250) calculates the observation probability value based on the observation probability value calculated by an observation probability calculator(230). A maximum similarity producer(260) calculates a local maximum similarity value based on the observation probability value.
Abstract translation: 目的：提供维特比解码器和用于识别语音的方法，以防止由非预期脉冲噪声引起的污染部分的观察概率的显着降低。 构成：最佳状态计算器（220）获得输入语音的观测向量行的每个测量向量中的最大累积相似度的状态。 缓冲单元（240）存储在输入的语音之前输入的多个语音的观察概率值。 非线性滤波单元250基于由观测概率计算器计算出的观测概率值来计算观测概率值。 最大相似度生成器（260）基于观察概率值计算局部最大相似度值。
-
公开(公告)号：KR1020100072751A
公开(公告)日：2010-07-01
申请号：KR1020080131243
申请日：2008-12-22
Applicant: 한국전자통신연구원
IPC: G10L21/0208 , G10L15/20
Abstract: PURPOSE: A method and an apparatus for reducing noises are provided to reinforce isolation function of voice and noise through voice/noise isolation function like soft masking technique thereby accurately presuming clean voice. CONSTITUTION: A noise estimator(130) presumes noise component within inputted voice signal. A posterior probability estimator(140) presumes posterior probability value from the noise component. A noise parameter adapting unit(150) applies noise Gaussian mixture model to the inputted voice signal. A voice/noise separating unit(160) divides noise and voice signal primarily. A noise removing unit(170) eliminates residual noise components of the voice signal.
Abstract translation: 目的：提供减少噪声的方法和装置，通过语音/噪声隔离功能（如软掩蔽技术）来加强语音和噪声的隔离功能，从而准确地推断干净的声音。 构成：噪声估计器（130）假设输入的语音信号内的噪声分量。 后验概率估计器（140）假设来自噪声分量的后验概率值。 噪声参数适应单元（150）将噪声高斯混合模型应用于输入的语音信号。 语音/噪声分离单元（160）主要分离噪声和语音信号。 噪声去除单元（170）消除语音信号的残余噪声分量。
-
公开(公告)号：KR1020090019967A
公开(公告)日：2009-02-26
申请号：KR1020070084301
申请日：2007-08-22
Applicant: 한국전자통신연구원
IPC: G10L21/0208 , G10L15/20 , G10L15/06
CPC classification number: G10L15/20
Abstract: An apparatus and a method for generating a noise adaptive acoustic model including discriminative noise adaptive training for environment transfer are provided to apply a voice recognition system to a noise environment effectively by using a voice recognition method. Voice studying data(201) reflects various noise environments by high-capacity audio data elementarily used for sound type model learning. A noise reduction unit(203) removes various noise components which are included in the voice studying data. A noise adaptive training technique unit(205) learns the voice studying data by the acoustic model training method. The learned acoustic model is set up as a basis acoustic model parameter(207). Voice data(211) for environmental adaptation is a small amount of voice data collected in the environment to which a voice recognition system is applied.
Abstract translation: 提供一种用于生成包括用于环境转移的辨别性噪声自适应训练的噪声自适应声学模型的装置和方法，以通过使用语音识别方法有效地将语音识别系统应用于噪声环境。 语音学习数据（201）通过大量音频数据反映各种噪声环境，这些音频数据是用于声型模型学习的基础。 噪声降低单元（203）去除包括在语音学习数据中的各种噪声分量。 噪声自适应训练技术单元（205）通过声学模型训练方法学习语音学习数据。 学习的声学模型被设置为基础声学模型参数（207）。 用于环境适应的语音数据（211）是在应用语音识别系统的环境中收集的少量语音数据。
-
公开(公告)号：KR100766094B1
公开(公告)日：2007-10-12
申请号：KR1020060094236
申请日：2006-09-27
Applicant: 한국전자통신연구원
Inventor： 강병옥
Abstract: A home network system based on a voice interface and a control method thereof are provided to integrally search multimedia contents dispersedly stored to various multimedia devices connected to a home network through a voice command and reproduce corresponding multimedia contents in a desired device without limitation of a location or a place, thereby maximizing user's convenience. A home network system based on a voice interface comprises a home media server. The home media server(100) comprises the followings: a profile manager module; a media transmission module(120); a device recognition and control module(130); a communication module(140); a voice processing module(160) which recognizes the voice command of a user, searches information about the recognized voice command, and generates and outputs a complex sound for a search result; and a central control unit(150) which stores multimedia contents information stored in a multimedia device connected to a home network, and controls that multimedia contents which the user wants.
Abstract translation: 提供了一种基于语音接口的家庭网络系统及其控制方法，通过语音命令对分散存储在各家多媒体设备上的多媒体内容进行综合搜索，并在所需设备中再现对应的多媒体内容，而不限于某个位置 或一个地方，从而最大限度地提高用户的便利性。 基于语音接口的家庭网络系统包括家庭媒体服务器。 家庭媒体服务器（100）包括：简档管理器模块; 媒体传输模块（120）; 设备识别和控制模块（130）; 通信模块（140）; 识别用户的语音命令的语音处理模块（160），搜索关于所识别的语音命令的信息，并生成并输出用于搜索结果的复合声音; 以及中央控制单元（150），其存储存储在连接到家庭网络的多媒体设备中的多媒体内容信息，并且控制用户想要的多媒体内容。
-
-
-
-
-
-
-
-
-
Search Results
133
records
(took 0.091 seconds)
