-
公开(公告)号：KR1020080090034A
公开(公告)日：2008-10-08
申请号：KR1020070032988
申请日：2007-04-03
IPC: G10L15/02
CPC classification number: G10L17/02
Abstract: A method and a system for recognizing a voice speaker are provided to accurately recognize a speaker and be resistant to a noise environment in recognizing a speaker. A voice detecting unit detects valid voice data from an inputted voice. A feature extracting unit extracts voice characteristics from the voice data. A feature converting unit(33) generates voice feature transform matrixes according to a PCA(Principle Component Analysis) and an LDA(Linear Discriminant Analysis) from the voice data, configures a hybrid voice feature transform matrix by combining the respective voice feature transform matrixes, and multiplies a matrix representing the voice feature to the hybrid voice feature conversion matrix to generate a final feature vector. A recognizing unit(35) generates a speaker model from the final feature vector to recognize a speaker in comparison with a previously stored general speaker model, and performs authentication on the recognized speaker.
Abstract translation: 提供用于识别语音扬声器的方法和系统，以便在识别扬声器时准确地识别扬声器并且抵抗噪声环境。 语音检测单元从输入的语音检测有效的语音数据。 特征提取单元从语音数据中提取语音特征。 特征转换单元（33）根据PCA（原理分量分析）和来自语音数据的LDA（线性判别分析）生成语音特征变换矩阵，通过组合各个语音特征变换矩阵来配置混合语音特征变换矩阵， 并将表示语音特征的矩阵与混合语音特征转换矩阵相乘以生成最终特征向量。 识别单元（35）从最终特征向量生成扬声器模型，以便与先前存储的一般扬声器模型相比较以识别扬声器，并对识别的扬声器进行认证。
-
Patent Agency Ranking
公开(公告)号：KR100926132B1
公开(公告)日：2009-11-11
申请号：KR1020070080679
申请日：2007-08-10
Applicant: 한국전자통신연구원
IPC: G01S3/808
Abstract: 본 발명은 임의로 배열된 세 개 이상의 마이크로폰(microphone)을 장착한 로봇 환경에서의 음원 방향 결정 기술에 관한 것으로, 특히 두 개의 마이크로폰 사이에서 각각 발생하는 신호 도달 시간 차이를 이용하여 음원의 방향을 탐지하는 기술에 관한 것이다. 본 발명에 따라 세 개의 마이크로폰을 사용할 경우, 총 여섯 개의 후보 방위각을 추정할 수 있는데, 이는 각각 해당 마이크로폰을 기준으로 한 방위각이다. 즉, 제 1 마이크로폰과 제 2 마이크로폰간의 시간 지연으로 추정된 두 개의 후보 방위각은 제 1 마이크로폰을 0°로 가정하여 추정된 각이고, 제 2 마이크로폰과 제 3 마이크로폰 간의 시간 지연으로 추정된 두 개의 후보 방위각은 제 2 마이크로폰을 0°로 가정하여 추정된 각이다. 따라서 여섯 개의 후보 방위각을 로봇의 정면을 기준으로 환산해야 한다. 각각의 마이크로폰 쌍에서 측정된 지연시간이 정확하다면 여섯 개의 각 중 세 개의 각은 군집해 있을 것이므로 이것을 음원의 방향으로 결정한다. 본 발명에 의하면, 임의로 배열된 세 개의 마이크로폰을 갖는 로봇에서 음원의 방향을 정확히 결정하도록 함으로써, 로봇이 사용자에게 보다 인간 친화적인 서비스를 제공할 수 있다.
지능형 로봇, 3채널, 음원 방향 탐지, 인간-로봇 상호작용-
公开(公告)号：KR1020090038697A
公开(公告)日：2009-04-21
申请号：KR1020070104131
申请日：2007-10-16
Applicant: 한국전자통신연구원
IPC: B25J13/08
Abstract: An intelligent robot for localizing sound source by frequency-domain characteristics and method thereof are provided to enable to call a robot with an improved sound tracking ability although the robot is far away. A method for localizing sound source of an intelligent robot comprises a step of receiving a voice signal through a multi-channel microphone and a sound source board(S200); a step of calculating delay time of each of sound signal(S206); a step of calculating a directional angle of the sound source using a retarding time information between the sound signals(S208); and a step of rotating a robot toward a directional angle of the sound source(S210). The voice signal is received through a multi-channel microphone and a sound source board.
Abstract translation: 提供了一种用于通过频域特性及其方法来定位声源的智能机器人，以便尽管机器人遥远，但能够以更好的声音跟踪能力来呼叫机器人。 一种用于定位智能机器人声源的方法包括通过多声道麦克风和声源板接收语音信号的步骤（S200）; 计算每个声音信号的延迟时间的步骤（S206）; 使用声音信号之间的延迟时间信息来计算声源的方位角的步骤（S208）; 以及使机器人朝向声源的方向角度旋转的步骤（S210）。 语音信号通过多声道麦克风和声源板接收。
-
公开(公告)号：KR1020090026879A
公开(公告)日：2009-03-16
申请号：KR1020070091933
申请日：2007-09-11
Applicant: 한국전자통신연구원
CPC classification number: G06K9/00234 , G06K9/00275 , G06K9/00288 , G06K9/44
Abstract: A feature extracting apparatus for face recognition based on a color image and a method thereof are provided to detect a peculiar face based on the color image and perform the face registration and recognition based on the detected peculiar face. A normalization part(100) normalizes plurality of RGB color images in a preset size. That is, it performs the inclination and size revision by using the pupil position information of a person in the color images, and then extracts an image of a preset size from the whole face image on the basis of the both eyes. After performing the linear conversion for the normalized respective RGB color images, a conversion unit(102) performs a histogram equalization process. An average image calculating unit(104) produces the average images for the plural images outputted from the conversion unit. A peculiar face generating unit(106) produces the covariance matrix of color images based on the average images calculated in the average image calculating unit, and the generates a peculiar face based on the calculated covariance matrix.
Abstract translation: 提供了一种基于彩色图像的用于人脸识别的特征提取装置及其方法，用于基于彩色图像检测特有的面部，并且基于检测到的特殊面进行面部登记和识别。 归一化部分（100）以预设大小归一化多个RGB彩色图像。 也就是说，通过使用彩色图像中的人的瞳孔位置信息来执行倾斜和尺寸修正，然后基于双眼从整个脸部图像中提取预设大小的图像。 在对归一化的各个RGB彩色图像执行线性转换之后，转换单元（102）执行直方图均衡处理。 平均图像计算单元（104）产生从转换单元输出的多个图像的平均图像。 特殊面部生成部（106）基于平均图像计算部中计算出的平均图像，生成彩色图像的协方差矩阵，根据所计算的协方差矩阵生成特殊面。
-
公开(公告)号：KR1020080023033A
公开(公告)日：2008-03-12
申请号：KR1020060087009
申请日：2006-09-08
Applicant: 한국전자통신연구원
Abstract: A voice recognition method using a wireless microphone in an intelligent robot service system and an apparatus thereof are provided to enable a speaker having a wireless microphone transmitter to replace a VAD(Voice Activity Detection) function with only on/off of the wireless microphone, thereby operating the voice recognition system as necessary and speaking freely without considering a distance from a robot or a speaking position. A voice recognition method comprises the following steps of: registering speakers by using a wireless microphone(201); receiving valid voice data through a wireless microphone receiver from a speaker which speaks through a wireless microphone transmitter(202,203); extracting a feature from the received valid voice data(203); generating at least one speaker model by using the extracted feature(204); and recognizing the speaker by measuring similarity between the extracted feature and the generated speaker model(205~207).
Abstract translation: 提供了一种在智能机器人服务系统中使用无线麦克风的语音识别方法及其装置，使得具有无线麦克风发送器的扬声器能够仅使用无线麦克风的开启/关闭来取代VAD（语音活动检测）功能，由此 在不考虑与机器人或说话位置的距离的情况下，根据需要和自由地操作语音识别系统。 语音识别方法包括以下步骤：通过使用无线麦克风来登记扬声器（201）; 通过无线麦克风接收器从通过无线麦克风发射器（202,203）通话的扬声器接收有效语音数据; 从所接收的有效语音数据（203）提取特征; 通过使用提取的特征（204）生成至少一个扬声器模型; 并且通过测量所提取的特征和所产生的说话者模型之间的相似度来识别说话者（205〜207）。
-
公开(公告)号：KR100776803B1
公开(公告)日：2007-11-19
申请号：KR1020060093539
申请日：2006-09-26
Applicant: 한국전자통신연구원
Abstract: An apparatus and a method for recognizing a speaker of an intelligence robot by using multi channel fuzzy fusion are provided to register a speaker in online through a multi channel microphone and recognize a voice in all directions of the robot, thereby exactly obtaining an omnidirectionally-spoken voice of the speaker with superior performance. A multi channel microphone unit(200) obtains and registers a voice of a speaker in online through plural channels. A voice data obtaining unit divides a sentence by detecting a start point and an end point regarding each registered voice by channel, and removes noise of the voice included in the divided sentence. A speaker model generating unit(400) constructs a speaker model based on characteristics for each voice data where the noise is removed, and performs conversion into a likelihood log value with respect to the constructed speaker model. A fuzzy processing unit(600) calculates a fused fuzzy value with respect to the likelihood log values by channel. A speaker recognizing unit(700) recognizes the maximum value of the calculated fuzzy value-based fusion values as a speaker.
Abstract translation: 提供了一种通过使用多通道模糊融合来识别智能机器人的扬声器的装置和方法，以通过多声道麦克风在线登记扬声器，并识别机器人的所有方向上的语音，从而准确地获得全方位的 扬声器的声音具有卓越的性能。 多声道麦克风单元（200）通过多个声道获取和登记扬声器在线的声音。 语音数据获取单元通过检测关于每个注册的声音的开始点和结束点来分割句子，并且消除分割语句中包含的语音的噪声。 扬声器模型生成单元（400）基于除去噪声的每个语音数据的特性构建扬声器模型，并且相对于所构建的扬声器模型执行转换为似然对数值。 模糊处理单元（600）根据频率计算出似然记录值的融合模糊值。 扬声器识别单元（700）将所计算的基于模糊值的融合值的最大值识别为扬声器。
-
-
-
-
-
Search Results
16
records
(took 0.02 seconds)
