基于T矩阵归一化PLDA的说话人确认

doi:10.3969/j.issn.1006-2475.2017.10.011

计算机与现代化

基于T矩阵归一化PLDA的说话人确认

兰州理工大学电气工程与信息工程学院，甘肃兰州730050

收稿日期:2017-03-07 出版日期:2017-10-30 发布日期:2017-10-31
作者简介:缑新科(1966-)，男，甘肃天水人，兰州理工大学电气工程与信息工程学院教授, 博士，研究方向：模式识别，信号处理；王跃（1990-），男，山东滕州人，硕士研究生，研究方向：模式识别，语音信号处理。

Speaker Verification of Normalization PLDA Based on T Matrix

College of Electrical and Information Engineering, Lanzhou University of Technology, Lanzhou 730050, China

Received:2017-03-07 Online:2017-10-30 Published:2017-10-31

摘要/Abstract

摘要： 利用i-vector/PLDA模型进行说话人确认时，对于不定时间的语音，由于将长度归一化后的i-vector转化到PLDA模型时，伴随着不确定性的扭曲和缩放，影响识别率。本文通过对全变量空间矩阵T的列向量执行归一化，代替在PLDA模型上对i-vector进行长度归一化，避免因在i-vector上执行长度归一化，导致转移到PLDA模型上产生不良的扭曲。实验结果表明，该方法得到和长度归一化相似的效果，部分效果要优于长度归一化。

关键词: i-vector/PLDA, 长度归一化, T矩阵, 高斯通用背景模型

Abstract: Recently, speaker verification based on i-vector/PLDA has become the state-of-the-art technique in speaker recognition.For the indefinite time speech, uncertainty of distortion and scaling, when i-vector with length normalization is converted to PLDA model, it affects the recognition rate. In this paper, the normalization of the length of the i-vector on the PLDA model is replaced by the normalization of total variability matrix T, to avoid the poor distortions. Experiments show that the method is similar to the length normalization, some of the results are better than that of the length normalization.

Key words: i-vector/PLDA, normalization of length, matrix T, GMM-UBM

缑新科，王跃. 基于T矩阵归一化PLDA的说话人确认[J]. 计算机与现代化, doi: 10.3969/j.issn.1006-2475.2017.10.011.

GOU Xin-ke, WANG Yue. Speaker Verification of Normalization PLDA Based on T Matrix[J]. Computer and Modernization, doi: 10.3969/j.issn.1006-2475.2017.10.011.

参考文献

［1］ Martin A F, Greenberg C S. The NIST 2010 speaker recognition evaluation［C］// The 11th Annual Conference of the International Speech Communication Association. 2010:2726-2729.

［2］ Kenny P, Stafylakis T, Ouellet P, et al. PLDA for speaker verification with utterances of arbitrary duration［C］// 2013 IEEE International Conference on Acoustics, Speech and Signal Processing. 2013:7649-7653.

［3］ Garcia-Romero D, Espy-Wilson C Y. Analysis of i-vector length normalization in speaker recognition systems［C］// Proceedings of the Annual Conference of the International Speech Communication Association. 2011:249-252.

［4］ Dehak N, Kenny P J, Dehak R, et al. Front-end factor analysis for speaker verification［J］. IEEE Transactions on Audio, Speech, and Language Processing, 2011, 19(4): 788-798.

［5］栗志意，张卫强，何亮,等.基于总体变化子空间自适应的 i-vector 说话人识别系统研究［J］. 自动化学报, 2014,40(8):1836-1840.

［6］ Reynolds D A, Quatieri T F, Dunn R B. Speaker verification using adapted Gaussian mixture models［J］. Digital Signal Processing, 2000,10(1):19-41.

［7］ Matějka P, Glembek O, Castaldo F, et al. Full-covariance UBM and heavy-tailed PLDA in i-vector speaker verification［C］// 2011 IEEE International Conference on Acoustics, Speech and Signal Processing(ICASSP). 2011:4828-4831.

［8］李琳,万丽虹,洪青阳,等. 基于概率修正PLDA的说话人识别系统［J］. 天津大学学报(自然科学与工程技术版), 2015,48(8):692-696.

［9］ Prince S J D, Elder J H. Probabilistic linear discriminant analysis for inferences about identity［C］// The 11th IEEE International Conference on Computer Vision. 2007:1-8.

［10］Sarkar A K, Matrouf D, Bousquet P M, et al. Study of the effect of i-vector modeling on short and mismatch utterance duration for speaker verification［C］// INTERSPEECH 2012. 2012:2662-2665.

［11］Stafylakis T, Kenny P, Ouellet P, et al. Text-dependent speaker recognition using PLDA with uncertainty propagation［C］// INTERSPEECH 2013. 2013:3684-3688.

基于T矩阵归一化PLDA的说话人确认

Speaker Verification of Normalization PLDA Based on T Matrix

可视化

摘要/Abstract

引用本文

使用本文

参考文献

相关文章 0

编辑推荐

Metrics

本文评价