Abstract
A transformation matrix linear interpolation (TMLI) approach for speaker adaptation is proposed. TMLI uses the transformation matrixes produced by MLLR from selected training speakers and the testing speaker. With only 3 adaptation sentences, the performance shows a 12.12% word error rate reduction. As the number of adaptation sentences increases, the performance saturates quickly. To improve the behavior of TMLI for large amounts of adaptation data, the TMLI+MAP method which combines TMLI with MAP technique is proposed. Experimental results show TMLI+MAP achieved better recognition accuracy than MAP and MLLR+MAP for both small and large amounts of adaptation data.
Similar content being viewed by others
References
Gauvain J L, Lee C H. Maximum a Posteriori Estimation for Multivariate Gaussian Mixture Observations of Markov Chains.IEEE Trans SAP, 1994,2 (2): 291–298.
Leggetter C J, Woodland P C. Maximum Likelihood Linear Regression for Speaker Adaptation of Continuous Density Hidden Markov Models.Computer Speech and Language, 1995,9 (2): 171–185.
Wang Z Y, Liu F. Speaker Adaptation Using Maximum Likelihood Model Interpolation,Proce ICASSP, 1999,2: 1368–1372.
Zhang J H, Tang X M.BAYES Methods. Changsha: National University of Defence Technology Publishing Company, 1993.
Xu T Z,The Analysis of Applied Fonctionelle, Beijing: Science Press, 2002.
Chang E, Zhou J L, Di S,et al. Large Vocabulary Mandarin Speech Recognition with Different Approaches in Modeling Tones.Proc ICSLP, 2000,2: 983–986.
Author information
Authors and Affiliations
Additional information
Foundation item: Supported by the Science and Technology Committee of Shanghai (01JC14033)
Biography: XU Xiang-hua (1977-), female, Ph. D. candidate, research direction: large vocabulary continuous Mandarin speech recognition and speaker adaptation
Rights and permissions
About this article
Cite this article
Xiang-hua, X., Jie, Z. Speaker adaptation with transformation matrix linear interpolation. Wuhan Univ. J. Nat. Sci. 9, 927–930 (2004). https://doi.org/10.1007/BF02850801
Received:
Issue Date:
DOI: https://doi.org/10.1007/BF02850801