公开(公告)号：KR1020130063767A

公开(公告)日：2013-06-17

申请号：KR1020110130309

申请日：2011-12-07

Applicant: 한국전자통신연구원

Inventor： 김상훈 ,  김경옥 ,  김인규 ,  주지호 ,  장기석

IPC: H01L27/146

CPC classification number: H01L31/1804 ,  G02B6/12004 ,  G02B6/131 ,  H01L31/02327 ,  H01L31/103 ,  H01L31/1037 ,  H01L31/109 ,  Y02E10/547 ,  Y02P70/521

Abstract: PURPOSE: A method for forming photodetectors is provided to simplify a manufacturing process by forming an optical waveguide, a first single crystal pattern, and a second single crystal pattern at the same time. CONSTITUTION: A buried insulating layer(110) and a semiconductor layer are formed on a substrate(100). A trench is formed in the semiconductor layer. A doping region(123) is formed in a part of the semiconductor layer. A first single crystal layer and a second single crystal layer are formed in the trench. The first single crystal layer, the second single crystal layer, and the semiconductor layer are patterned to form a first single crystal pattern(145), a second single crystal pattern(155), and an optical waveguide(125).

Abstract translation: 目的：提供一种用于形成光电检测器的方法，以通过同时形成光波导，第一单晶图案和第二单晶图案来简化制造过程。 构成：在基板（100）上形成掩埋绝缘层（110）和半导体层。 在半导体层中形成沟槽。 掺杂区域（123）形成在半导体层的一部分中。 在沟槽中形成第一单晶层和第二单晶层。 将第一单晶层，第二单晶层和半导体层图案化以形成第一单晶图案（145），第二单晶图案（155）和光波导（125）。

公开(公告)号：KR1020130059476A

公开(公告)日：2013-06-07

申请号：KR1020110125405

申请日：2011-11-28

Applicant: 한국전자통신연구원

Inventor： 김승희 ,  김동현 ,  김영익 ,  박준 ,  조훈영 ,  김상훈

IPC: G10L15/08

CPC classification number: G10L15/083 ,  G10L15/187 ,  G10L15/08 ,  G10L2015/081

Abstract: PURPOSE: A search space generating method for voice recognition and a system thereof are provided to improve an accuracy of a voice recognition by adding 'a pronunciation heat which is generated by a pronunciation conversion between recognition units' to a search space. CONSTITUTION: A WFST[Weighted Finite State Transducer] coupling unit generates a WFST L·G by a coupling of a WFST G[WFST Grammar] and a WFST L[WFST pronunciation Library] and generates a WFST L'·L·G by a coupling of a WFST L'[WFST pronunciation conversion] and the WFST L·G(310,320). The WFST coupling unit generates a WFST C·L'·L·G by a coupling of a WFST context[WFST C] and the WFST L'·L·G and generates a WFST H·C·L'·L·G by a coupling of a WFST H[WFST Hidden Markov model] and the WFST C·L'·L·G(330,340). A WFST optimization unit optimizes the WFST H·C·L'·L·G(350). [Reference numerals] (310) WFST G and WFST L combination; (320) WFST L' and WFST L·G combination; (330) WFST C and WFST L'·L·G combination; (340) WFST H and WFST C·L'·L·G combination; (350) Optimization; (AA) Start; (BB) End

Abstract translation: 目的：提供一种用于语音识别的搜索空间生成方法及其系统，以通过将由识别单元之间的语音转换产生的发音热量添加到搜索空间来提高语音识别的精度。 构成：WFST [加权有限状态传感器]耦合单元通过WFST G [WFST语法]和WFST L [WFST发音库]的耦合产生WFST L·G，并通过一个WFST L'·L·G生成WFST L' WFST L'[WFST发音转换]与WFST L·G（310,320）的耦合。 WFST耦合单元通过WFST上下文[WFST C]和WFST L'·L·G的耦合产生WFST C·L'·L·G，并通过下式产生WFST H·C·L'·L·G WFST H [WFST隐马尔可夫模型]和WFST C·L'·L·G（330,340）的耦合。 WFST优化单元优化WFST H·C·L'·L·G（350）。 （参考号）（310）WFST G和WFST L组合; （320）WFST L'和WFST L·G组合; （330）WFST C和WFST L'·L·G组合; （340）WFST H和WFST C·L'·L·G组合; （350）优化; （AA）开始; （BB）结束

公开(公告)号：KR1020120056086A

公开(公告)日：2012-06-01

申请号：KR1020100117611

申请日：2010-11-24

Applicant: 한국전자통신연구원

Inventor： 김동현 ,  김영익 ,  조훈영 ,  김승희 ,  김상훈 ,  박준

IPC: G10L15/14 ,  G10L15/06

CPC classification number: G10L15/14 ,  G10L15/26 ,  G10L19/038

Abstract: PURPOSE: An acoustic model adapting method and a voice recognizing device using the same are provided to eliminate a re-study burden of a user about a quantized acoustic model by an embedded voice recognizing machine. CONSTITUTION: An extracting unit(110) extracts features from a waveform corresponding to a voice. The extracting unit generates quantized data. A probability measuring unit(120) applies the quantized data, an adapted network, and a quantized acoustic model to fixed point-applied high-speed computation. The probability measuring unit calculates Gaussian occupancy probability. An adaption unit(130) updates the acoustic model. A voice recognizing unit(150) recognizes the extracted features using the updated acoustic model.

Abstract translation: 目的：提供一种声学模型适应方法和使用其的语音识别装置，以消除用户通过嵌入式语音识别机器对量化声学模型的重新学习负担。 构成：提取单元（110）从对应于声音的波形中提取特征。 提取单元生成量化数据。 概率测量单元（120）将量化数据，适应网络和量化声学模型应用于固定点施加的高速计算。 概率测量单元计算高斯占用概率。 适应单元（130）更新声学模型。 语音识别单元（150）使用更新的声学模型识别所提取的特征。

公开(公告)号：KR1020120004151A

公开(公告)日：2012-01-12

申请号：KR1020100064857

申请日：2010-07-06

Applicant: 한국전자통신연구원

Inventor： 김정세 ,  김상훈 ,  윤승 ,  이수종 ,  박상규

IPC: G06F17/28 ,  G06F17/30 ,  G10L15/04

CPC classification number: G06F17/289 ,  G10L15/26 ,  G10L25/78

Abstract: PURPOSE: A sentence translating device and a method thereof are provided to use morpheme information and pause information about a voice, thereby accurately separating a sentence. CONSTITUTION: A voice recognizing unit(20) generates a sentence of a first language based on a voice recognition result about a voice of the first language. A morpheme part-of-speech tagger(40) tags a morpheme part-of-speech from a sentence of the first language. A pause extracting unit(30) extracts pause information from the voice of the first language. A sentence separating unit(50) applies pause information whose length is longer than a threshold value when the sentence of the first language is separated.

Abstract translation: 目的：提供句子翻译装置及其方法，以使用语素信息和暂停关于语音的信息，从而准确地分离句子。 构成：语音识别单元（20）基于关于第一语言的语音的语音识别结果生成第一语言的句子。 词素标注器（40）的语素从第一语言的句子中标注语素部分。 暂停提取单元（30）从第一语言的语音中提取暂停信息。 当分离第一语言的句子时，句子分离单元（50）应用长度大于阈值的暂停信息。

公开(公告)号：KR1020110062547A

公开(公告)日：2011-06-10

申请号：KR1020090119308

申请日：2009-12-03

Applicant: 한국전자통신연구원

Inventor： 김상훈 ,  김경옥 ,  김인규 ,  서동우 ,  주지호 ,  장기석

IPC: G02B6/12 ,  H01L31/02

CPC classification number: G02B6/12004 ,  G02B2006/12176 ,  G02B2006/12188 ,  H01L31/105 ,  H01L31/1804 ,  H01L31/1812 ,  Y02E10/547 ,  Y02P70/521

Abstract: PURPOSE: An optical detector and a manufacturing method thereof are provided to simplify the process of forming a polycrystalline semiconductor layer, by growing the polycrystalline semiconductor layer from a single-crystalline semiconductor layer selectively. CONSTITUTION: A first single-crystalline semiconductor layer(121) and an optical waveguide(123) are formed. The optical waveguide is projected from the first single-crystalline semiconductor layer. An insulation layer(130,140) is formed on the first single-crystalline semiconductor layer. The insulation layer covers the optical waveguide. An opening(131) is formed by etching the insulation layer. The opening reveals the top surface of the optical waveguide. A second single-crystalline semiconductor layer(132) is formed in the opening. A polycrystalline semiconductor layer(133) doped with dopants is selectively formed, from the top surface of the second single-crystalline semiconductor layer.

Abstract translation: 目的：通过选择性地从单晶半导体层生长多晶半导体层，提供光学检测器及其制造方法，以简化形成多晶半导体层的工艺。 构成：形成第一单晶半导体层（121）和光波导（123）。 光波导从第一单晶半导体层突出。 在第一单晶半导体层上形成绝缘层（130,140）。 绝缘层覆盖光波导。 通过蚀刻绝缘层形成开口（131）。 开口显示光波导的顶表面。 在开口中形成第二单晶半导体层（132）。 从第二单晶半导体层的顶表面选择性地形成掺杂有掺杂剂的多晶半导体层（133）。

公开(公告)号：KR1020110017600A

公开(公告)日：2011-02-22

申请号：KR1020090075145

申请日：2009-08-14

Applicant: 한국전자통신연구원

Inventor： 김상훈 ,  박준 ,  박상규

IPC: G10L15/06 ,  G10L15/02 ,  G10L15/26

Abstract: PURPOSE: An apparatus for searching a word entry in a portable electronic dictionary and a method thereof are provided to output N-best recognition results and enables a user to select one of the results when performing a dictionary searching operation for a foreign language through a voice recognition technology. CONSTITUTION: A pre-treatment unit receives a voice signal for the combination of continuous pronunciation of each letter which configures a dictionary pronunciation or a word, and a word network configuration unit(512) receives a pronunciation string from the stored multi-pronunciation dictionary information to configure a network through the matching with the extracted phoneme series. A searching unit(514) refers to a triphone unit acoustic model transferred through a training unit and the configured network in order to search the word corresponding to the voice signal.

Abstract translation: 目的：提供一种用于搜索便携式电子词典中的词条的装置及其方法，用于输出N最佳识别结果，并且使用户能够通过语音执行外语的字典搜索操作时选择结果之一 识别技术。 构成：预处理单元接收用于组合字典发音或单词的每个字母的连续发音的组合的语音信号，并且字网络配置单元（512）从存储的多发音字典信息中接收发音串 通过与提取的音素系列匹配来配置网络。 搜索单元（514）是指通过训练单元和配置的网络传送的三音单元声学模型，以搜索对应于语音信号的单词。

公开(公告)号：KR1020100138520A

公开(公告)日：2010-12-31

申请号：KR1020090057093

申请日：2009-06-25

Applicant: 한국전자통신연구원

Inventor： 김정세 ,  김승희 ,  박준 ,  이수종 ,  윤승 ,  이일빈 ,  김상훈

IPC: G10L15/197 ,  G10L15/14 ,  G10L15/06

Abstract: PURPOSE: A speech recognition apparatus and a method thereof are provided to reduce error of remote speech recognition. CONSTITUTION: A syntax analyzing unit(23) analyzes syntax based on a morpheme word class to generate a hierarchical structure. A hierarchical word list generating unit(24) generates a word list by a hierarchy of a recognition word using the hierarchical structure. A hierarchical n-gram applying unit(25) generates a hierarchical n-gram score of the word list by a hierarchy. A calculation unit(27) adds the hierarchical n-gram score to sound and language model probability to generate a speech recognition score of the recognition word.

Abstract translation: 目的：提供语音识别装置及其方法，以减少远程语音识别的误差。 构成：语法分析单元（23）基于语素词类分析语法以生成层次结构。 分级词列表生成单元（24）使用层次结构通过识别词的层次来生成单词列表。 分层的n-gram应用单元（25）通过层级产生单词列表的分层n-gram分数。 计算单元（27）将分级n-gram分数与声音和语言模型概率相加，以产生识别词的语音识别分数。

公开(公告)号：KR1020100063607A

公开(公告)日：2010-06-11

申请号：KR1020090025685

申请日：2009-03-26

Applicant: 한국전자통신연구원

Inventor： 김상훈 ,  김경옥 ,  서동우 ,  주지호

IPC: C30B25/02 ,  C30B29/08 ,  H01L21/205 ,  H01L31/10

Abstract: PURPOSE: A growth method of a germanium single crystal thin film having a negative photoconductive property and an optical detector using the same are provided to improve a penetration dislocation density and a surface roughness by forming the germanium single crystal thin film of a high grade on a silicon substrate. CONSTITUTION: A germanium thin film is grown up(S11) on the silicon substrate in a low temperature. The germanium thin film is grown up by increasing temperature(S12). The germanium thin film is grown up(S13) in the high temperature. Each growth step is processed using a low pressure chemical vapor deposition. The deposition rate at the increasing temperature germanium growth is similar to the deposition rate at the low temperature germanium growth.

Abstract translation: 目的：提供具有负光导性的锗单晶薄膜和使用其的光学检测器的生长方法，以通过在高温下形成高等级的锗单晶薄膜来提高穿透位错密度和表面粗糙度 硅衬底。 构成：在低温下在硅衬底上长大（S11）锗薄膜。 锗薄膜通过增加温度而长大（S12）。 锗薄膜在高温下长大（S13）。 使用低压化学气相沉积处理每个生长步骤。 在增加温度锗生长时的沉积速率与低温锗生长时的沉积速度相似。

公开(公告)号：KR100924795B1

公开(公告)日：2009-11-03

申请号：KR1020070133391

申请日：2007-12-18

Applicant: 한국전자통신연구원

Inventor： 이수종 ,  박준 ,  김상훈 ,  이윤근 ,  김응규

IPC: G06T7/20 ,  G06K9/46 ,  G06T7/00

Abstract: 본 발명은 음성인식을 위해 수신되는 비디오를 분석하여 입술움직임이 있는지의 여부를 확인할 때, 다양한 움직임 영상을 대상으로 입술움직임 영상과 그 이외의 영상을 분류하는 입술움직임 영상 판별 방법 및 그 장치에 관한 것으로, 본 발명은 온라인 입술움직임 영상 판별 방법에 있어서, 촬영수단으로부터 수신되는 움직임영상프레임을 분석하여 입술움직임 영상에 대한 최종후보를 추출하는 제 1 단계; 및 영상추출수단으로부터 수신되는 상기 최종후보를 입술움직임 변별력 특징을 기준으로 입술움직임 영역과 여타요소 움직임 영역으로 온라인 상에서 레이블링하고, 최종후보 중에서 입술움직임 영역 및 여타요소 움직임 영역으로 분류되지 않은 최종후보에 대한 입술움직임 영상 여부를 SVM 영역분류구분선을 근거로 판별하는 제 2 단계;를 포함하는 것을 특징으로 한다.
SVM 패턴분류, 입술움직임 영상 판별

公开(公告)号：KR100820141B1

公开(公告)日：2008-04-08

申请号：KR1020060064262

申请日：2006-07-10

Applicant: 한국전자통신연구원

Inventor： 이수종 ,  김상훈 ,  이영직 ,  김응규

IPC: G10L25/78 ,  G10L15/25

Abstract: 본 발명은 음향 수신부와 영상 수신부가 구비된 음성 구간 검출 장치에 있어서, 상기 영상 수신부로부터 출력되는 영상 프레임에서 움직임 영역을 검출하고, 상기 검출된 움직임 영역에 입술 움직임 영상 특징 정보를 적용하여 입술 움직임 신호를 검출하는 입술 움직임 신호 검출부, 상기 음향 수신부로부터 출력되는 음향 프레임과 상기 입술 움직임 신호 검출부에서 검출된 입술 움직임 신호를 이용하여 음성 구간을 검출하는 음성 구간 검출부를 포함하는 것으로서, 음성구간 검출과정에서 입술움직임 영상정보를 확인하기 때문에 dynamic 잡음이 음성으로 오인식 되는 것을 미리 방지할 수 있다.
음성구간, 음성인식, 입술움직임