特征空间本征音说话人自适应

屈丹; 杨绪魁; 张文林

doi:10.16383/j.aas.2015.c140644

特征空间本征音说话人自适应

doi: 10.16383/j.aas.2015.c140644

1.
解放军信息工程大学信息系统工程学院郑州 450000

基金项目:

国家自然科学基金(61175017, 61403415, 61302107)资助

详细信息

作者简介:
杨绪魁中国人民解放军信息工程大学信息系统工程学院博士研究生. 主要研究方向为语音信号处理, 语音识别.E-mail: gzyangxk@163.com

计量
- 文章访问数: 2032
- HTML全文浏览量: 127
- PDF下载量: 1257
- 被引次数: 0
出版历程
- 收稿日期: 2014-09-12
- 修回日期: 2015-01-24
- 刊出日期: 2015-07-20

Feature Space Eigenvoice Speaker Adaptation

1.
Institute of Information Systems Engineering, Information Engineering University, Zhengzhou 450000

Funds:

Supported by National Natural Science Foundation of China (61175017, 61403415, 61302107)

摘要

摘要: 提出了特征空间本征音说话人自适应算法,该方法首先借鉴RATZ 算法的思想,采用高斯混合模型对特征空间中的说话人信息进行建模;其次利用子空间方法实现对特征补偿项的估计,减少估计参数的数量,在对特征空间精确建模的同时,降低了算法对自适应数据量的需求.基于微软语料库的中文连续语音识别实验表明,该算法在自适应数据量极少时仍能取得较好的性能,配合说话人自适应训练能够进一步降低词错误率,其实时性优于本征音说话人自适应算法.
- 连续语音识别 /
- 说话人自适应 /
- 多高斯倒谱规整 /
- 本征音
Abstract: A speaker adaptation method at feature level named feature-space eigenvoice adaptation method is proposed. In this method, similar to RATZ, the information of speakers in the feature space is modeled by a Gaussian mixture model. Moreover, the number of parameters to be estimated is decreased by taking the dependency of these parameters into account. This method can use very little data to construct a more accurate feature space model. Experimental results of continuous speech recognition on Microsoft speech database show that this method can still achieve good performance even when the adaptation data is limited. And speaker adaptive training based on this method can further decrease the word error rate with a superior real-time performance to that of eigenvoice methods.
- Continuous speech recognition /
- speaker adaptation /
- multivariate Gaussian-based cepstral normalization /
- eigenvoice

HTML全文

参考文献(19)

[1]	Teng W X, Gravier G, Bimbot F, Soufflet F. Speaker adaptation by variable reference model subspace and application to large vocabulary speech recognition. In: Proceedings of the 2009 IEEE International Conference on Acoustics, Speech and Signal Processing. Taiwan, China: IEEE, 2009. 4381-4384
[2]	Zhang W L, Zhang W Q, Li B C, Qu D, Johnson M T. Bayesian speaker adaptation based on a new hierarchical probabilistic model. IEEE Transactions on Audio, Speech, and Language Processing, 2012, 20(7): 2002-2015
[3]	Zhang W L, Qu D, Zhang W Q, Li B C. Rapid speaker adaptation using compressive sensing. Speech Communication, 2013, 55(10): 950-963
[4]	Kenny P, Boulianne G, Dumouchel P. Eigenvoice modeling with sparse training data. IEEE Transactions on Speech and Audio Processing, 2005, 13(3): 345-354
[5]	Varadarajan B, Povey D, Chu S M. Quick FMLLR for speaker adaptation in speech recognition. In: Proceedings of the 2008 International Conference on Acoustics, Speech, and Signal Processing. Las Vegas, Nevada, USA: IEEE, 2008. 4297-4300
[6]	Ghoshal A, Povey D, Agarwal M, Akyazi P, Burget L, Kai Feng, Glembek O, Goel N, Karafiat M, Rastrow A, Rose R C, Schwarz P, Thomas S. A novel estimation of feature-space MLLR for full-covariance models. In: Proceedings of the 2010 IEEE International Conference on Acoustics, Speech, and Signal Processing. Dallas, TX, USA: IEEE, 2010. 4310-4313
[7]	Rath S P, Povey D, Vesely K, Cernocky J. Improved feature processing for Deep Neural Networks. In: Proceedings of the 14th Annual Conference of the International Speech Communication Association, Lyon, France: ISCA, 2013. 109-113
[8]	Rath S P, Burget L, Karafiát M, Glembek O, Cernocký J. A region-specific feature-space transformation for speaker adaptation and singularity analysis of Jacobian matrix. In: Proceedings of the 2013 Annual Conference of International Speech Communication Association. Lyon, France: ISCA, 2013. 1228-1232
[9]	Ghalehjegh S H, Rose R C. Two-stage speaker adaptation in subspace gaussian mixture models. In: Proceedings of the 2014 IEEE International Conference on Acoustics, Speech and Signal Processing. Florence, Italy: IEEE, 2014. 6324-6328
[10]	Chen S, Kingsbury B, Mangu L, Povey D, Saon G, Soltau H, Zweig G. Advances in speech transcription at IBM under the DARPA EARS program. IEEE Transactions on Audio, Speech, and Language Processing, 2006, 14(5): 1596-1608
[11]	Saon G, Chien J T. Large-vocabulary continuous speech recognition systems: a look at some recent advances. IEEE Signal Processing Magazine, 2012, 29(6): 18-33
[12]	Joshi V, Prasad V N, Umesh S. Modified cepstral mean normalization-transforming to utterance specific non-zero mean. In: Proceedings of the 2013 Annual Conference of International Speech Communication Association. Lyon, France: ISCA, 2013. 881-885
[13]	Buera L, Lleida E, Miguel A, Ortega A, Saz O. Cepstral vector normalization based on stereo data for robust speech recognition. IEEE Transactions on Audio, Speech, and Language Processing, 2007, 15(3): 1098-1113
[14]	Droppo J, Deng L, Acero A. Uncertainty decoding with SPLICE for noise robust speech recognition. In: Proceedings of the 2002 IEEE International Conference on Acoustics, Speech, and Signal Processing. Orlando, FL, USA: IEEE, 2002. I-57-I-60
[15]	Moreno P J, Raj B, Stern R M. Data-driven environmental compensation for speech recognition: a unified approach. Speech Communication, 1998, 24(4): 267-285
[16]	Wang Y Q, Gales M J F. Model-based approaches to adaptive training in reverberant environments. In: Proceedings of the 2012 Annual Conference of International Speech Communication Association. Portland, Oregon: ISCA, 2012. 959-963
[17]	Ochiai T, Matsuda S, Lu X G, Hori C, Katagiri S. Speaker adaptive training using deep neural networks. In: Proceedings of the 2014 IEEE International Conference on Acoustics, Speech, and Signal Processing. Florence, Italy: IEEE, 2014. 6349-6353
[18]	Povey D, Ghoshal A, Boulianne G, Burget L, Glembek O, Goel N, Hannemann M, Motlicek P, Qian YM, Schwarz P, Silovsky J, Stemmer G, Vesely K. The Kaldi speech recognition toolkit. In: Proceedings of the 2011 IEEE Workshop on Automatic Speech Recognition and Understanding. Hawaii, USA: IEEE, 2011.
[19]	Eric C, Zhou J L, Shi Y, Huang C. Speech lab in a box: a Mandarin speech toolbox to jumpstart speech related research. In: Proceedings of the 2001 European Conference on Speech Communication and Technology. Scandinavia, Aalborg, Denmark: ISCA, 2001. 2799-2782