期刊文献+

误差可控的GMM似然率近似算法

下载PDF
导出
摘要 在语音识别领域,连续概率密度GMM广泛应用于声学建模.由于标准算法复杂度高,各种基于高斯选择的似然率近似算法被提出,包括:部分距离消去(PDE)、最优混合分量预测(BMP)算法,以及部分距离消去的动态高斯选择算法(DGS).本文通过对各种算法的分析比较,在DGS的基础上引入误差控制参数,在保证算法运行效率的同时,有效的控制了运算误差的大小,为解决似然率近似算法中误差不可控的问题提供了新的思路.实验结果显示,此算法在不降低识别率的前提下,识别时间较标准算法减少10%以上.
出处 《漳州师范学院学报(自然科学版)》 2013年第1期33-39,共7页 Journal of ZhangZhou Teachers College(Natural Science)
  • 相关文献

参考文献12

  • 1X. Huang, A. Acero, and H.-W. Hon. Spoken Language Processing: A Guide to Theory Algorithm and System Development(1st edition) [M]. Prentice Hall PTR, New Jersy, 2001.
  • 2S. Kanthac, K. Schutz, and H. Ney, "Using S1MD instructions for fast likelihood calculation in LVCSR," in Proceedings ofICASSP\2000,3: 1531-1534.
  • 3G F. G Yared, F. Violaro, and L. C. Sousa. Gaussian Elimination Algorithm for HMM Complexity Reduction in ContinuousSpeech Recognition Systems[J]. in Proceedings of inter speech, 2005, 377-380.
  • 4M. J. F. Gales,K. M. Knill, S. J. Young. State-Based Gaussian Selection in Large VocabularyContinuous Speech RecognitionUsing HMM's [J]. IEEE Transactions on Speech and Audio Processing, 1999, 7(2): 152-161.
  • 5E. Bocchieri. Vector Quantization for the Efficient Computation of Continuous Density Likelihood[J]. Proceedings ofICASSP '93,1993, 2: 692-695.
  • 6Jun Cai,Ghazi Bouselmi, Dominique Fohr, Yves L.aprie. Dynamic Gaussian Selection Technique for Speeding up HMM-basedContinuous Speech Recognition[D]. In Proceedings of 2008 IEEE International Conference on Acoustics, Speech, and SignalProcessing (ICASSP 2008), March 30 , April 4,2008,Las Vegas.
  • 7B. L. Pellom, R. Sarikaya, J. H. L. Hansen. Fast Likelihood Computation Techniques in Nearest-Neighbor Based Search forContinuous Speech Recognition [J]. IEEE Signal Processing Letters, 2001,8(8): 221-224.
  • 8Kai-Fu Lee, Hsiao-Wuen Hon. Speaker Independent Phone Recognition Using Hidden Markvo Models [J], IEEE Transaction OnAcoustics Speech and Signal Processing, 1989,37( 11): 1641-1648.
  • 9R. Schwartz, Y. Chow, 0. Kimball, S. Roucos, M. Krasner, and J.Makhoul. Context-dependent modeling for acoustic-phoneticrecognition of continuous speech[J]. presented at the IEEE Int. Conf. Acoust" Speech, Signal Processing, 1985.
  • 10Lori F.Lamel and Jean-Luc Gauvain. Experiments on Speaker-Independent Phone Recognition Using BREF[J]. Acoustics,Speech, and Signal Processing, 1992. ICASSP-92., 1992 IEEE International Conference on Acoust" Speech, Signal Processing,1992, H 23-26): 557-560.

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