期刊文献+

基于PLDA的多信道多语音说话人确认研究 被引量:3

PLDA for Speaker Verification under Multi-Channel and Multi-Record
下载PDF
导出
摘要 在NIST SRE 2012年评测和实际应用中,可以用说话人的多个语音样本来注册说话人模型,并且这些语音样本取自于各种各样的信道。本文基于PLDA,尝试了多种打分方法,并提出一种新的得分规整技术,在NIST SRE 2012核心测试集上,EER平均提升26.0%,MinCost平均提升12.4%。 In NIST SRE 2012 evaluation and practical applications,multiple recordings,which come from various channel conditions, can be used to train a speaker model. Based on PLDA,this paper will try several score methods and propose one score normalization technique. Equal error rate and minimum cost has been relatively improved 26. 0% and 12. 4% respectively on NIST SRE 2012 core test corpus.
出处 《网络新媒体技术》 2014年第1期13-19,共7页 Network New Media Technology
基金 国家自然科学基金(批准号:10925419 90920302 61072124 11074275 11161140319 91120001 61271426) 中国科学院战略性先导科技专项(面向感知中国的新一代信息技术研究 编号:XDA06030100 XDA06030500) 国家863计划(资助号:2012AA012503) 中科院重点部署项目(编号:KGZD-EW-103-2)经费资助
关键词 说话人识别 PLDA 多语音 得分规整 speaker recognition PLDA multi-record score normalization
  • 相关文献

参考文献14

  • 1N Dehak,P Kenny,R Dehak. Front-End Factor Analysis For Speaker Verification[J].IEEE Transactions on Audio Speech and Language Processing,2011,(04):788-798.
  • 2P Kenny. Bayesian speaker verification with heavy tailed Priors[A].Brno,Czech Rebublic,2010.
  • 3N Brummer. EM for Probabilistic LDA[OL].https://sites.google.com/site/nikobrummer,2010.
  • 4M Senoussaoui,P Kenny,N Brummer. Mixture of PLDA models in i-vector space for gender independent speaker recognition[A].Florence,Italy,2011.
  • 5P Matejka,O Glembek,F Castaldo. Full-covariance UBM and heavy-tailed PLDA in i-vector speaker verification[A].Prague,Czech Republic,2011.4536-4539.
  • 6L Burget,O Plchot,S Cumani. Discriminatively trained probabilistic linear discriminant analysis for speaker verification[A].Prague,Czech Republic,2011.4832-4835.
  • 7N Dehak,R Dehak,J Glass. Cosine similarity scoring without score normalization techniques[A].Brno,Czech Rebublic,2010.
  • 8S Cumani,N Brummer,L Burget. Fast discriminative speaker verification in the i-vector space[A].Prague,Czech Republic,2011.4852-4855.
  • 9J Villalba,N Brummer. Towards fully Bayesian speaker recognition:Integrating out the between speaker covariance[A].Florence,Italy,2011.
  • 10T Stafylakis,P Kenny,M M Senoussaoui. Preliminary investigation of Boltzmann machine classifiers for speaker recognition[A].Biopolis,Singapore,2012.

同被引文献9

引证文献3

二级引证文献4

相关作者

内容加载中请稍等...

相关机构

内容加载中请稍等...

相关主题

内容加载中请稍等...

浏览历史

内容加载中请稍等...
;
使用帮助 返回顶部