高级搜索

留言板

尊敬的读者、作者、审稿人, 关于本刊的投稿、审稿、编辑和出版的任何问题, 您可以本页添加留言。我们将尽快给您答复。谢谢您的支持!

姓名
邮箱
手机号码
标题
留言内容
验证码

基于K-L散度的最大后验弧主导的混淆网络生成算法

王欢良 韩纪庆 郑铁然 李海峰

王欢良, 韩纪庆, 郑铁然, 李海峰. 基于K-L散度的最大后验弧主导的混淆网络生成算法[J]. 电子与信息学报, 2008, 30(5): 1109-1112. doi: 10.3724/SP.J.1146.2006.01760
引用本文: 王欢良, 韩纪庆, 郑铁然, 李海峰. 基于K-L散度的最大后验弧主导的混淆网络生成算法[J]. 电子与信息学报, 2008, 30(5): 1109-1112. doi: 10.3724/SP.J.1146.2006.01760
Wang Huan-liang, Han Ji-qing, Zheng Tie-ran, Li Hai-feng . K-L Divergence based Confusion Network Generation Algorithm Guided with Maximum Posteriori Arc[J]. Journal of Electronics & Information Technology, 2008, 30(5): 1109-1112. doi: 10.3724/SP.J.1146.2006.01760
Citation: Wang Huan-liang, Han Ji-qing, Zheng Tie-ran, Li Hai-feng . K-L Divergence based Confusion Network Generation Algorithm Guided with Maximum Posteriori Arc[J]. Journal of Electronics & Information Technology, 2008, 30(5): 1109-1112. doi: 10.3724/SP.J.1146.2006.01760

基于K-L散度的最大后验弧主导的混淆网络生成算法

doi: 10.3724/SP.J.1146.2006.01760
基金项目: 

国家自然科学基金(60575030)和黑龙江省留学回国基金(LC03C10)资助课题

K-L Divergence based Confusion Network Generation Algorithm Guided with Maximum Posteriori Arc

  • 摘要: 为快速生成高质量混淆网络,该文提出一种最大后验弧主导的快速生成算法。它只需遍历一遍Lattice,具有线性时间复杂度。采用K-L散度(Kullback-Leibler Divergence,KLD)来度量弧标号之间的发音相似性,改善了混淆网络生成中弧对齐的准确性。实验结果显示,所提算法在生成速度上和Xue的快速算法是可比的,而生成质量更好。通过采用KLD作为弧标号相似性测度,生成混淆网络的质量得到了进一步提高。
  • Mangu L, Brill E, and Stolcke A. Finding consensus in speechrecognition: Word error minimization and other applicationsof confusion networks[J].Computer Speech and Language.2000,14(4):373-400[2]Tur G, Wright J, and Gorin A, et al.. Improving spokenlanguage understanding using word confusion networks.Proceedings of ICSLP, Denver, Colorado, 2002: 1137-1140.[3]Bertoldi N and Federico M. A new decoder for spokenlanguage translation based on confusion networks. IEEEASRU Workshop, Cancun, Mexico, 2005: 134-140.[4]Xue J and Zhao Y X. Random forests-based confidenceannotation using novel feature from confusion network.Proceedings of ICASSP, Toulouse, France, 2006: 1149-1152.Hillard D and Ostendorf M. Compensation forward posteriorestimation bias in confusion networks. Proceedings ofICASSP, Toulouse, France, 2006: 1153-1156.[5]Hakkani-Tur D and Riccardi G. A general algorithm for wordgraph matrix decomposition. Proceedings of ICASSP, HongKong, China, 2003: 596-599.[6]Xue J and Zhao Y X. Improving confusion network algorithmand shortest path search from word lattice. Proceedings ofICASSP, Philadelphia, PA, 2005: 853-856.[7]Kullback S and Leibler R A. On information and sufficiency[J].Ann. Math. Stat.1951, 22(1):79-86[8]Liu P, Soong F K, and Zhou J L. Effective estimation ofKullback-Leibler divergence between speech models. Tech.Rep., Microsoft Research Asia, 2005.[9]Chang E, Shi Y, and Zhou J L, et al.. Speech lab in a box: aMandarin speech toolbox to Jumpstart speech relatedresearch. Proceedings of Eurospeech, Aalborg, Denmark,2001: 2799-2802.
  • 加载中
计量
  • 文章访问数:  3745
  • HTML全文浏览量:  85
  • PDF下载量:  903
  • 被引次数: 0
出版历程
  • 收稿日期:  2006-11-09
  • 修回日期:  2007-07-02
  • 刊出日期:  2008-05-19

目录

    /

    返回文章
    返回