-
1.
公开(公告)号:KR1020120040649A
公开(公告)日:2012-04-27
申请号:KR1020110098935
申请日:2011-09-29
Applicant: 삼성전자주식회사 , 서울대학교산학협력단
CPC classification number: G10L19/265 , G10L15/02 , G10L15/28
Abstract: PURPOSE: A pre processing device for voice recognition, a device thereof, and a method thereof are provided to convert the voice of a test environment by a linear dynamic system, thereby increasing the recognition rate of the voice recognition device. CONSTITUTION: A voice input unit divides an first input voice to a fixed frame(S10). A voice converting unit applies conversion rules to the frame of the first voice. The voice converting unit converts the frame of the first voice to a frame of a second voice(S20). A recognizing unit identifies verbal meaning by recognizing frames of the second voice(S30).
Abstract translation: 目的:提供一种用于语音识别的预处理装置,其装置及其方法,用于通过线性动态系统转换测试环境的语音,从而增加语音识别装置的识别率。 构成:语音输入单元将第一输入语音划分为固定帧(S10)。 语音转换单元将转换规则应用于第一语音的帧。 语音转换单元将第一语音的帧转换为第二语音的帧(S20)。 识别单元通过识别第二语音的帧来识别语言意义(S30)。
-
公开(公告)号:KR101862352B1
公开(公告)日:2018-05-30
申请号:KR1020110098935
申请日:2011-09-29
Applicant: 삼성전자주식회사 , 서울대학교산학협력단
Abstract: (a) 음성인식장치에입력되는제1음성을소정의프레임으로분할하는단계; (b) 상기분할된각각의프레임에변환규칙을적용하여상기제1음성의프레임을제2음성의프레임으로변환하는단계; 및 (c) 상기음성인식장치가상기변환된제2음성의프레임을인식하는단계를포함하되, 상기 (b) 단계는, 상기제1음성의프레임의이전에위치한프레임들중 적어도하나를반영하여상기제1음성의프레임을상기제2음성의프레임으로변환하는단계를포함하는본 발명의일 실시예에따른음성인식방법이개시된다.
-
公开(公告)号:KR101145441B1
公开(公告)日:2012-05-15
申请号:KR1020110036463
申请日:2011-04-20
Applicant: 서울대학교산학협력단
Abstract: PURPOSE: A method for combining sounds of a statistical speech combining system using a switching linear dynamic system is provided to apply a voice database and a switching linear dynamic system which learns an existing education system, thereby combining voices. CONSTITUTION: A system learns a statistical model from a voice database. The system learns a system parameter of a switching linear dynamic system using an ML(Maximum Likelihood) method(S100). The system selects a statistical model and a converter corresponding to an input sentence or an input word. The system uses a learnt statistical model value as an input value. The system combines voices which are from a combined feature vector of the switching linear dynamic system(S200).
Abstract translation: 目的:提供一种使用切换线性动态系统组合统计语音组合系统的声音的方法,以应用语音数据库和学习现有教育系统的切换线性动态系统,从而组合语音。 构成:系统从语音数据库学习统计模型。 该系统使用ML(最大似然)方法(S100)学习切换线性动态系统的系统参数。 系统选择对应于输入语句或输入单词的统计模型和转换器。 系统使用学习的统计模型值作为输入值。 该系统组合了来自开关线性动态系统(S200)的组合特征向量的语音。
-
-