-
公开(公告)号：KR1020100070097A
公开(公告)日：2010-06-25
申请号：KR1020080128695
申请日：2008-12-17
Applicant: 한국전자통신연구원
Abstract: PURPOSE: A method for processing a hybrid feature vector, a method and an apparatus for recognizing a speaker using the same are provided to perform a speaker recognition service regardless of noises by combining a multi-streaming method and a sub-band recombination method. CONSTITUTION: A first feature vector generation unit generates a first feature vector based on an input voice. A second feature vector generation unit time-filters the first feature vector to generate a second feature vector. The first and the second feature vector processing units(150a, 150b) calculate similarities of sub-bands of the first and the second feature vectors. A hybrid unit(107) merges and outputs the similarity.
Abstract translation: 目的：提供一种用于处理混合特征向量的方法，用于识别使用该混合特征向量的扬声器的方法和装置，以通过组合多流传输方法和子带复合方法来执行讲话者识别服务而不管噪声如何。 构成：第一特征矢量生成单元基于输入声音生成第一特征向量。 第二特征向量生成单元对第一特征向量进行时间滤波以生成第二特征向量。 第一和第二特征向量处理单元（150a，150b）计算第一和第二特征向量的子带的相似度。 混合单元（107）合并并输出相似度。
-
公开(公告)号：KR101059892B1
公开(公告)日：2011-08-29
申请号：KR1020080114116
申请日：2008-11-17
Applicant: 한국전자통신연구원
Abstract: 본 발명은 개개의 화자식별 결과의 신뢰도 측정 방법에 관한 것으로, 각 프레임의 화자식별 결과에 대한 공헌 정도를 측정하고, 이 각 프레임의 화자식별 공헌도를 기반으로 화자식별 결과의 신뢰도를 측정하고, 이를 화자 진위 판단에 이용함으로써, 화자 검증시에 제시된 화자의 진위를 정확하게 판단하여, 원거리 다채널 환경에서 화자식별의 정확도를 높일 수 있다.
화자식별, 화자식별 결과의 신뢰도, 원거리 다채널 환경, 화자식별 결과 통합.Abstract translation: 本发明测量测量的程度基于每个帧结果的扬声器识别的贡献以促进与所述个体说话者识别结果的可靠性测量方法的每个帧的说话者识别结果，识别所述说话者的可靠性，它 可以准确地确定在验证讲话者时呈现的讲话者的真实性，从而提高远程多信道环境中讲话者识别的准确性。
-
公开(公告)号：KR1020100055168A
公开(公告)日：2010-05-26
申请号：KR1020080114116
申请日：2008-11-17
Applicant: 한국전자통신연구원
Abstract: PURPOSE: A speaker identification method for a multi-channel speaker identification system is provided to improve the performance of speaker identification and speaker verification in a remote distance multi-channel environment. CONSTITUTION: A speaker identifier for a multi-channel speaker identification system is connected to each input microphone(101). A speaker identification integrator(115) integrates the identification result of the speaker identifier. The channel speaker identifier computes the degree of speaker identification contribution by the measurement of posterior probability of a registered speaker. The speaker identification integrator produces channel reliability through speaker identification contribution. The speaker identification integrator integrates the channel speaker identification result with the weight application by the channel reliability.
Abstract translation: 目的：提供多声道扬声器识别系统的扬声器识别方法，以提高远程多声道环境中扬声器识别和扬声器验证的性能。 构成：用于多声道扬声器识别系统的扬声器标识符连接到每个输入麦克风（101）。 扬声器识别积分器（115）集成了扬声器标识符的识别结果。 信道扬声器标识符通过测量注册说话者的后验概率来计算扬声器识别贡献的程度。 扬声器识别积分器通过扬声器识别贡献产生通道可靠性。 扬声器识别积分器通过信道可靠性将声道扬声器识别结果与重量应用相结合。
-
公开(公告)号：KR101081972B1
公开(公告)日：2011-11-09
申请号：KR1020080128695
申请日：2008-12-17
Applicant: 한국전자통신연구원
Abstract: 본발명은하이브리드특징벡터를이용하여화자인식의정확성을향상시키는방법에관한것으로서, 특징벡터 AS-MFCC(Autocorrelation Sequence - Mel-Frequency Cepstral Coefficient)와 RAS-MFCC(Relative AS-MFCC)를동시에사용하는멀티스트리밍방법과서브밴드특징벡터재결합방법을혼합하여이용하며, 본발명에의하면잡음환경에서정확하고신뢰도높은인식결과를제공할수 있다.
-
-
-
Search Results
4
records
(took 0.015 seconds)
