公开(公告)号：KR100366703B1

公开(公告)日：2003-01-09

申请号：KR1020000010876

申请日：2000-03-04

Applicant: 삼성전자주식회사

Inventor： 차선화 ,  최인정 ,  공병구

IPC: G10L15/22

Abstract: PURPOSE: A speech recognition apparatus of a human response type is provided to get exactly solved indistinct of sound and meaning in speech recognition. CONSTITUTION: A speech characteristic extraction part(10) converts an analog speech signal into a digital data and extracts speech characteristics. A model searching part(20) applies a sound model and a language model into the extracted speech characteristics and produces one more sentences considered for recognition. A sentence unit indistinct discriminating part(30) discriminates whether to exist indistinct in a sentence unit among the sentences considered for recognition. An indistinct section extract part extracts an indistinct section of the sentences. A question producing part(50) generates questions for canceling the extracted indistinct section. A user response processing part(60) analyzes user responses and extracts response corresponding to the indistinct section so as to cancel the indistinct of the sentences.

Abstract translation: 目的：提供一种人类响应类型的语音识别装置，以准确解决语音识别中声音和含义的模糊。 构成：语音特征提取部分（10）将模拟语音信号转换成数字数据并提取语音特征。 模型搜索部分（20）将声音模型和语言模型应用到提取的语音特征中，并产生一个考虑用于识别的句子。 句子单元不清楚鉴别部分（30）鉴别在被考虑识别的句子中是否以句子为单位存在不明确。 不清楚的部分提取部分提取句子的不清楚的部分。 问题产生部分（50）产生用于取消提取的不清楚部分的问题。 用户响应处理部分（60）分析用户响应并提取对应于不清楚部分的响应以消除句子的不清楚。

公开(公告)号：KR1020010086902A

公开(公告)日：2001-09-15

申请号：KR1020000010876

申请日：2000-03-04

Applicant: 삼성전자주식회사

Inventor： 차선화 ,  최인정 ,  공병구

IPC: G10L15/22

Abstract: PURPOSE: A speech recognition apparatus of a human response type is provided to get exactly solved indistinct of sound and meaning in speech recognition. CONSTITUTION: A speech characteristic extraction part(10) converts an analog speech signal into a digital data and extracts speech characteristics. A model searching part(20) applies a sound model and a language model into the extracted speech characteristics and produces one more sentences considered for recognition. A sentence unit indistinct discriminating part(30) discriminates whether to exist indistinct in a sentence unit among the sentences considered for recognition. An indistinct section extract part extracts an indistinct section of the sentences. A question producing part(50) generates questions for canceling the extracted indistinct section. A user response processing part(60) analyzes user responses and extracts response corresponding to the indistinct section so as to cancel the indistinct of the sentences.

Abstract translation: 目的：提供人类反应类型的语音识别装置，以便在语音识别中精确解决声音和意义上的不确定性。 构成：语音特征提取部（10）将模拟语音信号转换为数字数据并提取语音特征。 模型搜索部分（20）将声音模型和语言模型应用于提取的语音特征中，并产生多于一个被认为用于识别的句子。 句子单位不清楚的辨别部分（30）判断在被认为是识别的句子中句子单位是否存在模糊。 一个不清楚的部分提取部分提取句子的模糊部分。 问题产生部分（50）产生用于取消提取的模糊部分的问题。 用户响应处理部分（60）分析用户响应并提取与模糊部分相对应的响应，以消除句子的模糊。