公开(公告)号：KR100463376B1

公开(公告)日：2004-12-29

申请号：KR1020020078215

申请日：2002-12-10

Applicant: 한국전자통신연구원

Inventor： 유초롱 ,  윤승 ,  최미란 ,  오승신 ,  박준 ,  이영직

IPC: G10L13/00

Abstract: PURPOSE: A translation engine device for translating a source language into an object language and a translation method therefor are provided to use a dialogue sentence in several domain environment and accurately translate the source language inputted by a user into the object language. CONSTITUTION: A mapping table(408) stores a cluster of an object language mapped with a cluster of a source language. A DTST(Direct Translation Sentence Table) direct translation unit(401) directly translates a sentence capable of direct translation in an inputted source language sentence. A pre-processing module(402) maintains a kernel language of the source language sentence through the morpheme analysis of the source language sentence, hides the other portions, and simplifies the structure of the sentence. A clustering unit(404) divides the source language sentence into clusters. A mapping unit(405) decides the cluster of the object language mapped to the cluster of the source language using the mapping table(408). A post-processing and generating unit(406) reallocates the order of the clusters of the object language, and recovers the object language as a completed sentence form.

Abstract translation: 目的：提供一种翻译引擎装置，用于将源语言翻译成目标语言及其翻译方法，以使用多个领域环境中的对话语句并将用户输入的源语言准确地翻译成目标语言。 构成：映射表（408）存储与源语言的聚类映射的对象语言的聚类。 DTST（直接翻译句子表）直接翻译单元（401）直接翻译输入的源语言句子中能够直接翻译的句子。 预处理模块（402）通过源语言句子的词素分析来维护源语言句子的核心语言，隐藏其他部分，并且简化句子的结构。 聚类单元（404）将源语言句子分为聚类。 映射单元（405）使用映射表来决定映射到源语言的聚类的目标语言的聚类（408）。 后处理和生成单元（406）重新分配对象语言的聚类的顺序，并且将对象语言恢复为完整的句子形式。

公开(公告)号：KR1020080045413A

公开(公告)日：2008-05-23

申请号：KR1020060114504

申请日：2006-11-20

Applicant: 한국전자통신연구원

Inventor： 김상훈 ,  오승신

IPC: G10L13/08 ,  G10L15/14 ,  G10L13/06

Abstract: A break predicting method to which a static feature and a dynamic feature are reflected, a text-to-speech system based on the same, and a method therefor are provided to combine a CART(Classification And Regression Tree) model of the static feature with an HMM(Hidden Markov Model) model of the dynamic feature, generate a new break prediction model, and predict the most corresponding break strength to the meaning of the corresponding sentence through the generated break prediction model. Text data are extracted from a text corpus(S210). Morphological analysis for the extracted text data is performed(S230). A feature parameter is extracted from the morphological analysis result(S240). The voice recording of the extracted text data is performed, and training data are configured(S250). CART modeling is performed on the basis of the training data, and observation probability is calculated(S260). HMM modeling is performed on the basis of the training data, and transition probability is calculated(S270). A break prediction model is generated on the basis of the observation probability and the transition probability(S280). If a sentence is inputted, a break strength for the inputted sentence is predicted through the break prediction model.

Abstract translation: 一种反映静态特征和动态特征的突破预测方法，基于该静态特征和动态特征的文本到语音系统及其方法被提供以将静态特征的CART（分类和回归树）模型与 动态特征的HMM（隐马尔可夫模型）模型，生成新的断点预测模型，并通过生成的突变预测模型预测最相应的断裂强度与相应句子的含义。 从文本语料库提取文本数据（S210）。 执行提取的文本数据的形态分析（S230）。 从形态分析结果中提取特征参数（S240）。 执行提取的文本数据的语音记录，并配置训练数据（S250）。 基于训练数据进行CART建模，计算观察概率（S260）。 基于训练数据进行HMM建模，计算转移概率（S270）。 基于观测概率和转移概率生成中断预测模型（S280）。 如果输入了句子，则通过中断预测模型来预测输入的句子的断裂强度。

公开(公告)号：KR100806287B1

公开(公告)日：2008-02-22

申请号：KR1020060072502

申请日：2006-08-01

Applicant: 한국전자통신연구원

Inventor： 오승신 ,  김상훈

IPC: G10L13/08

Abstract: 본 발명은 문말 억양 예측 방법 및 이를 기반으로 하는 음성합성 방법 및 시스템에 관한 것으로, 문말 억양과 가장 밀접한 상관 관계를 가지는 문장의 종결어미를 이용하여 문말 억양 예측 모델을 생성하고 생성된 문말 억양 예측 모델을 통해 입력된 대화체 문장의 의미에 가장 부합하는 문말 억양을 생성함으로써, 보다 자연스러운 대화체 합성음을 생성할 수 있는 것을 특징으로 한다.
음성합성시스템(Text-to-Speech system), 문말 억양(sentence-final intonation), 양태(modality), 운율(prosody), 화행(speech act), 문형(sentence type)

公开(公告)号：KR1020080011859A

公开(公告)日：2008-02-11

申请号：KR1020060072502

申请日：2006-08-01

Applicant: 한국전자통신연구원

Inventor： 오승신 ,  김상훈

IPC: G10L13/08

Abstract: A method for predicting a sentence-final intonation and a Text-to-Speech method and system based on the same are provided to generate the sentence-final intonation by using the final ending of a sentence with a strong correlation between the sentence-final intonation and the final ending of the sentence, thereby generating the sentence-final intonation most suitable for the meaning of a conversational sentence and generating a more natural conversational complex sound. A method for predicting a sentence-final intonation comprises the following steps of: extracting text data in consideration of the distribution of a final ending type from a conversational text corpus; performing sentence-final intonation tagging for the extracted text data based on a sentence-final intonation tag set which is set by a sentence-final intonation type(S520); performing aspect tagging for the extracted text data based on an aspect tagging table that the final ending of a sentence is divided by meanings(S530); performing speech-act tagging and sentence pattern tagging for the extracted text data(S540); constructing training data based on the text data in which the sentence-final intonation tagging, aspect tagging, speech-act tagging, and sentence pattern tagging are completed; generating a sentence-final intonation prediction model for sentence-final intonation prediction by a statistical method based on the training data; and performing the sentence-final intonation tagging by predicting the sentence-final intonation for a conversational sentence through the sentence-final intonation prediction model when the conversational sentence is inputted(S550,S560).

Abstract translation: 提供了一种用于预测句子最终语调的方法和基于其的文本到语音方法和系统，以通过使用句子最终语调之间强相关性的句子的最终结尾来生成句子最终语调 并最终结束句子，从而产生最适合于会话句子意义的句子最终语调，并产生更自然的对话复合声音。 一种用于预测句子最后语调的方法包括以下步骤：从会话文本语料库中考虑最终结束类型的分布来提取文本数据; 基于由句子最终语调类型设置的句子最终语调标签集（S520），对所提取的文本数据执行句子最终语调标记; 基于通过句子的最终结尾被划分的方面的标记表来执行所提取的文本数据的方面标记（S530）; 对所提取的文本数据执行语音行为标记和句型标记（S540）; 根据完成语句最后语调标签，方面标记​​，言语行为标签和句型标签的文本数据构建训练数据; 通过基于训练数据的统计学方法生成句子最终语调预测的句子最终语调预测模型; 并且通过在输入会话语句时通过句子最终语调预测模型预测会话句子的句子最终语调（S550，S560）来执行句子最终语调标签。

公开(公告)号：KR100669241B1

公开(公告)日：2007-01-15

申请号：KR1020040106610

申请日：2004-12-15

Applicant: 한국전자통신연구원

Inventor： 오승신 ,  김종진 ,  최문옥 ,  이영직 ,  김상훈

IPC: G10L13/10

CPC classification number: G10L13/08 ,  G10L13/04

Abstract: 본 발명은 화행 정보를 이용한 대화체 음성합성 시스템 및 방법에 관한 것으로서, 대화 텍스트(dialog text)에서 대화의 맥락(context)에 따라 다른 억양이 구현될 필요가 있는 표현에 대해 두 대화자의 발화 문장으로부터 추출되는 화행(speech act) 정보를 이용하여 억양을 구분하는 태깅을 수행해 주고, 음성 합성시에는 그 태그에 맞는 억양을 갖는 음성 신호를 음성데이타베이스에서 추출하여 합성에 사용함으로써 대화의 흐름에 맞는 자연스럽고 다양한 억양을 구현함으로써, 대화의 상호작용(interaction)적인 측면을 좀더 실감나게 표현할 수 있어 대화음성의 자연성의 증진 효과를 기대할 수 있다.
대화체 음성합성시스템(Dialog-style Text-to-Speech system), 대화체 텍스트(dialog text), 음성 합성(speech synthesis), 맥락(context), 화행(speech act), 억양(intonation)

公开(公告)号：KR100644814B1

公开(公告)日：2006-11-14

申请号：KR1020050106584

申请日：2005-11-08

Applicant: 한국전자통신연구원

Inventor： 오승신 ,  김상훈 ,  이영직

IPC: G10L13/033 ,  G10L13/10

Abstract: A device and a method for synthesizing voices are provided to realize various styles of voices with a voice database of a singular radio performer, thereby vividly expressing conversation voices. Levels of intimacy are defined(S10). Voices recording text constructed corresponding to each intimacy level are stored(S20). At least one of a sentence final intonation contour pattern, an intonation pattern of a primary intonation phrase in a sentence, and a pitch mean value of a sentence of each voice data is statistically modeled to extract a metrical characteristic according to each intimacy(S30). Rhythm models by intimacy levels are generated based on the extracted metrical characteristic(S40).

Abstract translation: 提供了一种用于合成语音的设备和方法，以利用单一无线电演员的语音数据库来实现各种类型的语音，由此生动地表达会话语音。 亲密度水平被定义（S10）。 存储对应于每个亲密度等级构造的记录文本的声音（S20）。 的至少一个的句末语调轮廓图案的，伯语调短语的句子中的一个语调模式，以及基音意味着每个语音数据的一个句子的值统计学建模根据每个亲密提取韵律特性（S30） 。 基于提取的度量特性生成亲密度级别的节奏模型（S40）。

公开(公告)号：KR1020040050394A

公开(公告)日：2004-06-16

申请号：KR1020020078215

申请日：2002-12-10

Applicant: 한국전자통신연구원

Inventor： 유초롱 ,  윤승 ,  최미란 ,  오승신 ,  박준 ,  이영직

IPC: G10L13/00

Abstract: PURPOSE: A translation engine device for translating a source language into an object language and a translation method therefor are provided to use a dialogue sentence in several domain environment and accurately translate the source language inputted by a user into the object language. CONSTITUTION: A mapping table(408) stores a cluster of an object language mapped with a cluster of a source language. A DTST(Direct Translation Sentence Table) direct translation unit(401) directly translates a sentence capable of direct translation in an inputted source language sentence. A pre-processing module(402) maintains a kernel language of the source language sentence through the morpheme analysis of the source language sentence, hides the other portions, and simplifies the structure of the sentence. A clustering unit(404) divides the source language sentence into clusters. A mapping unit(405) decides the cluster of the object language mapped to the cluster of the source language using the mapping table(408). A post-processing and generating unit(406) reallocates the order of the clusters of the object language, and recovers the object language as a completed sentence form.

Abstract translation: 目的：提供用于将源语言翻译成对象语言的翻译引擎设备及其翻译方法，以在多个域环境中使用对话语句，并将用户输入的源语言准确地翻译成对象语言。 构成：映射表（408）存储与源语言的集群映射的对象语言的集群。 DTST（直译翻译句）直接翻译单元（401）直接翻译能够在输入的源语言句中直接翻译的句子。 预处理模块（402）通过源语句的语素分析维护源语言句的内核语言，隐藏其他部分，并简化了句子的结构。 聚类单元（404）将源语句划分成簇。 映射单元（405）使用映射表（408）来确定映射到源语言的集群的对象语言的集群。 后处理和生成单元（406）重新分配对象语言的簇的顺序，并将对象语言恢复为完整的句子形式。