基于I-Vector的多核学习SVM的说话人确认系统  

Speaker verification system of multiple-kernel-learning SVM based on I-Vector

作　　者：龚铖[1] 琚炜 

机构地区：[1]中国科学技术大学信息科学技术学院,安徽合肥230026

出　　处：《微型机与应用》2017年第22期15-18,22,共5页Microcomputer & Its Applications

摘　　要：自I-Vector(身份认证矢量)被提出以来,基于I-Vector的说话人确认系统迅速取代了基于GMM超矢量的系统并开始流行。I-Vector-SVM系统作为其中之一,在通常训练样本较少的说话人确认领域有着独特的优势,但其性能受核函数影响较大。因此,基于多核学习(Multiple Kernel Learning,MKL)思想,构建了基于I-Vector的多核学习SVM说话人确认系统,并与I-VectorSVM基线系统进行了性能比较。基于NIST语料库的实验表明,基于I-Vector的多核学习说话人确认系统相对于基线系统可取得一定的性能提升。bAs the concept ‘I-Vector＇was put forward,the text-independent speaker verification systems based on GMM super vector was replaced by the same systems based on I-Vector. As one of the systems,I-Vector-SVM system has a potential advantage when facing a small amount of training data. But its performance is influenced by its kernel too much. Under this situation,this paper builds a MKL-SVM speaker verification system based on I-Vector inspired by the concept‘multiple kernel learning＇,and compares it with the I-Vector-SVM baseline system. The experiment result based on NIST database showed,this system has an advantage in performance comparing with the baseline system.

关 键 词：说话人确认 多核学习SVM I-Vector 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]