|本期目录/Table of Contents|

[1]吴文昭.基于GMM聚类的鲁棒性i向量说话人确认[J].工业仪表与自动化装置,2017,(04):55-59.
　WU Wenzhao.Speaker verification robust speaker recognition based on i-vector and GMM clustering[J].Industrial Instrumentation & Automation,2017,(04):55-59.
点击复制

基于GMM聚类的鲁棒性i向量说话人确认

《工业仪表与自动化装置》[ISSN:1000-0682/CN:61-1121/TH]

卷:
期数:
2017年04期
页码:
55-59
栏目:
出版日期:
2017-08-15

文章信息/Info

Title:
Speaker verification robust speaker recognition based on i-vector and GMM clustering
文章编号:
1000-0682（2017）04-0000-00
作者:
吴文昭
（兰州城市学院 信息工程学院，兰州 730000）
Author(s):
WU Wenzhao
(School of Information Engineering, Lanzhou City University, Lanzhou 730000，China)
关键词:
说话人识别; 高斯混合模型; 巴氏距离; 支持向量机; 线性判别分析
Keywords:
speaker recognition;  Gaussian mixture model;  Bhattacharyya distance;  support vector machine;  linear discriminant analysis
分类号:
TP391
DOI:
-
文献标志码:
A
摘要:
针对i向量说话人确认系统识别率低且鲁棒性差的问题，提出一种基于GMM聚类的鲁棒性i向量生成算法，应用于SVM说话人识别系统。该算法根据话者GMM模型间的巴氏距离，对说话者GMM模型进行聚类，将N个说话人模型划分为K类，再根据聚类中心模型，应用MAP算法提取聚类超向量，采用联合因子分析方法提取其i向量，对得到的i向量应用线性判别分析和类内协方差归一化技术对其进行信道补偿和降维。将该i向量用于训练SVM以判定目标说话人，仿真实验验证了该算法的有效性。
Abstract:
For the sake of improving system performance and robustness in speaker verification based on i-vector, a novel feature extraction method based on GMM clustering was proposed in this paper. In the method, the paper clustered GMMs according to Bhattacharyya distance between pairs of GMMs. By doing so, the input GMMs were divided into K categories. This method extracted GMM clustering super-vectors based on clustering center models using MAP. Immediately following, i-vectors were extracted using Joint factor analysis. In the phase of restrain channel interference, we proposed a sequential channel compensation approach of LDA followed by WCCN. SVM was adopted as classifier to decide the target speaker. Experimental results verified the effectiveness of our algorithm.

参考文献/References:

[1] 忻栋,杨莹春,吴朝晖.基于SVM-HMM混合模型的说话人确认[J].计算机辅助设计与图形学学报,2002,14(11): 1080-1082.

[2] Vapnik V N, Vapnik VN. An overview of statistical learning theory. IEEE Trans Neural Netw 10:988-999[J]. IEEE Transactions on Neural Networks,1999,10(5): 988-99.
[3] N Dehak. Front-end factor analysis for speaker verification [J].IEEE Transactions on Audio,Speech and Language Processing. 2011, 19(4): 788-798.
[4] Kenny P, Boulianne G, Ouellet P, et al. Joint Factor Analysis Versus Eigenchannels in Speaker Recognition[J]. IEEE Transactions on Audio Speech & Language Processing, 2007, 15(4):1435-1447.
[5] Garcia-Romero D, Espy-Wilson C Y. Analysis of i-vector Length Normalization in Speaker Recognition Systems[C]// Florence,Italy: INTERSPEECH 2011, Conference of the International Speech Communication Association, 2011: 3283-3291.
[6] 栗志意,张卫强,何亮,等.基于核函数的IVEC-SVM说话人识别系统研究[J].自动化学报,2014(4):780-784.
[7] NIST. The NIST Year 2010 Speaker Recognition Evaluation Plan[OL].available:http://www.nist.gov/speech/tests/sre/2010/index.html, September 12, 2012.
[8] 谭萍,邢玉娟.基于GMM超向量和Fisher-稀疏表示分类的说话人确认[J].青海大学学报(自然科学版),2016,34(1): 51-57.
[9] 陈霄鹏,彭亚雄,贺松.基于PLDA的说话人识别时变鲁棒性问题研究[J].微型机与应用,2016,35(5):58-60.
[10] 高新建,李弼程,屈丹.基于类内方差归一化和SVM的说话人识别方法[J].计算机工程与应用,2009,45(10): 168-171.
[11] 于娴,贺松,彭亚雄,等.基于GMM模型的声纹识别模式匹配研究[J].通信技术,2015,48(1):97-101.
[12] 刘靖明,韩丽川,侯立文.基于粒子群的K均值聚类算法[J].系统工程理论与实践,2005,25(6):54-58.
[13] 胡伟.改进的层次K均值聚类算法[J].计算机工程与应用, 2013,49(2):157-159.
[14] 宣国荣,郑俊翔,杨程云,等.巴氏距离和K-L变换结合的特征选择[J].计算机工程与应用,2004,40(36):90-92.
[15] Flores-Sintas A, Cadenas J M, Martin F. Detecting homogeneous groups in clustering using the Euclidean distance[J].Fuzzy Sets & Systems,2001,120(120):213-225.

相似文献/References:

[1]邢玉娟,谭 萍.基于稀疏表示分类的说话人识别算法及其在智能考勤系统中的应用[J].工业仪表与自动化装置,2016,(02):84.
　XING Yujuan,TAN Ping.Speaker recognition based on sparse representation applied in intelligent attendance system[J].Industrial Instrumentation & Automation,2016,(04):84.
[2]叶 珍,白 璘.局部保护降维与高斯混合模型的高光谱图像分类[J].工业仪表与自动化装置,2017,(04):3.
　YE Zhen,BAI Lin.Hyperspectral image classification based on locality-preserving dimension reduction and Gaussian mixture model[J].Industrial Instrumentation & Automation,2017,(04):3.

备注/Memo

备注/Memo:
收稿日期：2016-11-21
基金项目：甘肃省教育厅科研项目（2015B-090）
作者简介：吴文昭（1966），甘肃天水人，副教授，硕士研究生，主要研究方向为生物特征识别。
更新日期/Last Update: 1900-01-01