公开(公告)号：KR100449912B1

公开(公告)日：2004-09-22

申请号：KR1020020008979

申请日：2002-02-20

Applicant: 대한민국(전남대학교총장)

Inventor： 김진영 ,  최승호 ,  이경록

IPC: G10L15/187

Abstract: PURPOSE: A postprocessing method for detecting a keyword of a voice recognition system is provided to improve discrimination for verifying validity of keywords, thereby preventing erroneously detected keywords from being admitted. CONSTITUTION: Voice data is inputted to a keyword detector(S10). The keyword detector detects keywords from the input voice data for obtaining interval information and probability of the detected keywords(S20). A phoneme recognizer analyzes the keyword interval information for deciding an interval of each phoneme and calculates the phoneme unit probability at intervals corresponding to the phonemes(S30). A phoneme interval estimator defines a phoneme model by the phoneme interval information decided by a word recognizer, defines a half-phoneme model by the phoneme interval information decided by the phoneme recognizer, and estimates phoneme intervals based on the phoneme interval information(S40). A similarity comparator calculates a phoneme model probable value and a half-phoneme probable value in each phoneme interval, and calculates an ACM(Anti-filler Confidence Measure) capable of comparing the similarity of the phoneme model and the half-phoneme model(S50). A validity discriminator compares the ACM with a threshold(S60). If the ACM exceeds the threshold, the validity discriminator admits the validity of the keyword(S70). If the ACM is lower than the threshold, the validity discriminator does not admit the validity of the keyword(S80).

Abstract translation: 目的：提供一种用于检测语音识别系统的关键词的后处理方法，以改善用于验证关键字的有效性的鉴别，由此防止错误地检测到的关键词被接纳。 构成：将声音数据输入到关键字检测器（S10）。 关键字检测器从输入语音数据中检测关键字以获得检测到的关键字的间隔信息​​和概率（S20）。 音素识别器分析用于确定每个音素的间隔的关键字间隔信息，并以对应于这些音素的间隔计算音素单位概率（S30）。 音素间隔估计器通过由单词识别器决定的音素间隔信息来定义音素模型，通过由音素识别器决定的音素间隔信息来定义半音素模型，并且基于音素间隔信息估计音素间隔（S40）。 相似度比较器计算每个音素间隔中的音素模型可能值和半音素可能值，并且计算能够比较音素模型和半音素模型的相似度的ACM（反填充置信度测量）（S50） 。 有效性鉴别器将ACM与阈值进行比较（S60）。 如果ACM超过阈值，则有效性鉴别器承认关键字的有效性（S70）。 如果ACM低于阈值，则有效性鉴别器不承认关键字的有效性（S80）。

公开(公告)号：KR1020040075447A

公开(公告)日：2004-08-30

申请号：KR1020030010946

申请日：2003-02-21

Applicant: 대한민국(전남대학교총장)

Inventor： 김진영 ,  최승호

IPC: H04W4/18

Abstract: PURPOSE: A mobile communication-based voice recognition system and method are provided to prevent the performance of voice recognition from being deteriorating due to distorted voice in a communication channel. CONSTITUTION: An exchange(200) allows a call connection and data transmission between a mobile terminal(100) and a voice recognition server(300). The voice recognition server(300) provides a service to a user according to a voice recognition performance result by a post-processing unit(310). A pattern matching unit(311) obtains a matching probability or a distance of each word by comparing a specific vector column received from the mobile terminal(100) with a word model stored in a word model database(312). An acknowledgement determining unit(313) selects a word with the lowest probability or a word with the closest distance, verifies whether the recognized word is an actually pronounced word, and outputs a corresponding result.

Abstract translation: 目的：提供一种基于移动通信的语音识别系统和方法，以防止语音识别的性能由于通信信道中的语音失真而恶化。 构成：交换机（200）允许移动终端（100）和语音识别服务器（300）之间的呼叫连接和数据传输。 语音识别服务器（300）根据后处理单元（310）的语音识别性能结果向用户提供服务。 模式匹配单元（311）通过将从移动终端（100）接收的特定向量列与存储在单词模型数据库（312）中的单词模型进行比较来获得每个单词的匹配概率或距离。 确认确定单元（313）选择具有最低概率的字或具有最近距离的字，验证所识别的字是否是实际发音的字，并输出相应的结果。

公开(公告)号：KR1020030069377A

公开(公告)日：2003-08-27

申请号：KR1020020008978

申请日：2002-02-20

Applicant: 대한민국(전남대학교총장)

Inventor： 김진영 ,  최승호 ,  이경록

IPC: G10L15/02

Abstract: PURPOSE: An apparatus and a method for detecting a topic of a voice recognition system are provided to increase discrimination by increasing the number of components of a basic analysis unit, thereby reducing an error of topic detection. CONSTITUTION: If voice data is inputted to an apparatus for detecting a topic, a phoneme recognizer(100) recognizes phonemes from the input voice data. A key string detector(110) separately packs the recognized phonemes for integrating the phonemes by the base analysis unit. A topic comparator(120) calculates the probability of the basic analysis unit that the basic analysis unit appears from topic data of trained data preliminarily memorized, and judges appropriateness of each topic by the probability of appearance for granting a score. A topic detector(130) detects topics more than a preliminarily set threshold in the score-granted topics.

Abstract translation: 目的：提供一种用于检测语音识别系统的主题的装置和方法，以通过增加基本分析单元的分量的数量来增加鉴别，从而减少主题检测的错误。 构成：如果语音数据被输入到用于检测主题的装置，则音素识别器（100）从输入的语音数据识别音素。 钥匙串检测器（110）分别包装用于由基本分析单元对音素进行积分的识别的音素。 主题比较器（120）根据预先记录的训练数据的主题数据计算基本分析单元出现的基本分析单元的概率，并根据出现评分的概率判断每个主题的适当性。 主题检测器（130）在得分授予主题中检测超过预设阈值的主题。

公开(公告)号：KR1020030069378A

公开(公告)日：2003-08-27

申请号：KR1020020008979

申请日：2002-02-20

Applicant: 대한민국(전남대학교총장)

Inventor： 김진영 ,  최승호 ,  이경록

IPC: G10L15/187

Abstract: PURPOSE: A postprocessing method for detecting a keyword of a voice recognition system is provided to improve discrimination for verifying validity of keywords, thereby preventing erroneously detected keywords from being admitted. CONSTITUTION: Voice data is inputted to a keyword detector(S10). The keyword detector detects keywords from the input voice data for obtaining interval information and probability of the detected keywords(S20). A phoneme recognizer analyzes the keyword interval information for deciding an interval of each phoneme and calculates the phoneme unit probability at intervals corresponding to the phonemes(S30). A phoneme interval estimator defines a phoneme model by the phoneme interval information decided by a word recognizer, defines a half-phoneme model by the phoneme interval information decided by the phoneme recognizer, and estimates phoneme intervals based on the phoneme interval information(S40). A similarity comparator calculates a phoneme model probable value and a half-phoneme probable value in each phoneme interval, and calculates an ACM(Anti-filler Confidence Measure) capable of comparing the similarity of the phoneme model and the half-phoneme model(S50). A validity discriminator compares the ACM with a threshold(S60). If the ACM exceeds the threshold, the validity discriminator admits the validity of the keyword(S70). If the ACM is lower than the threshold, the validity discriminator does not admit the validity of the keyword(S80).

Abstract translation: 目的：提供一种用于检测语音识别系统的关键字的后处理方法，以改善用于验证关键词的有效性的歧视，从而防止错误检测到的关键字被允许。 构成：将语音数据输入到关键字检测器（S10）。 关键词检测器检测来自输入语音数据的关键字，以获得间隔信息和检测到的关键词的概率（S20）。 音素识别器分析用于确定每个音素的间隔的关键词间隔信息，并且以对应于音素的间隔计算音素单位概率（S30）。 音素间隔估计器通过由字识别器确定的音素间隔信息定义音素模型，通过由音素识别器确定的音素间隔信息定义半音素模型，并且基于音素间隔信息来估计音素间隔（S40）。 相似度比较器计算每个音素间隔中的音素模型可能值和半音可能值，并且计算能够比较音素模型和半音素模型的相似度的ACM（防止填充物置信度测量）（S50） 。 有效性鉴别器将ACM与阈值进行比较（S60）。 如果ACM超过阈值，则有效性鉴别器承认关键字的有效性（S70）。 如果ACM低于阈值，则有效性鉴别器不承认关键字的有效性（S80）。