Abstract
Both HMMs and n-gram models are usually created by estimating their parameters on some sample set. Afterwards, the trained models can be applied for the segmentation of new data. This is by definition not part of the training samples and can never be in practical applications. Thus, the characteristic properties of this test data can be predicted to a limited extent only on the basis of the training material. Therefore, in general differences between training and testing material will occur that cannot be captured by the statistical models created. Ultimately, this mismatch between training and testing conditions will adversely affect the quality of the results achieved.
Therefore, it is the common goal of model adaptation to compensate differences between the training and testing conditions of a recognition system which concern the statistical properties of the data. In this chapter, the most important techniques will be presented that have been proposed for the adaptation of HMMs and n-gram models.
This is a preview of subscription content, log in via an institution to check access.
Access this chapter
Tax calculation will be finalised at checkout
Purchases are for personal use only
Notes
- 1.
This effect can be partially avoided in practice by a suitable tying of parameters (cf. Sect. 9.2, p. 169).
- 2.
By means of suitable extensions of the MLLR method, the parameters of covariance matrices can be adapted, too [107]. However, the rather moderate improvements observed in practice do usually not justify the considerably increased additional effort.
- 3.
If all mean vectors are adapted by a single affine transformation only, a single regression class comprising all densities is used.
- 4.
Here c(z|w 1,…,w n ) denotes the number of occurrences of word z in the string of words given by w 1,…,w n .
References
Bigi, B., De Mori, R., El-Béze, M., Spriet, T.: Combined models for topic spotting and topic-dependent language modeling. In: Furui, S., Huang, B.H., Chu, W. (eds.) Proc. Workshop on Automatic Speech Recognition and Understanding, pp. 535–542 (1997)
Boulis, C., Diakoloukas, V.D., Digalakis, V.V.: Maximum likelihood stochastic transformation adaptation for medium and small data sets. Computer Speech & Language 15, 257–285 (2001)
Boulis, C., Digalakis, V.V.: Fast speaker adaptation of large vocabulary continuous density HMM speech recognizer using a basis transform approach. In: Proc. Int. Conf. on Acoustics, Speech, and Signal Processing, Istanbul, pp. 3630–3633 (2000)
Clarkson, P.R., Robinson, A.J.: Language model adaptation using mixtures and an exponentially decaying cache. In: Proc. Int. Conf. on Acoustics, Speech, and Signal Processing, München, vol. 2, pp. 799–802 (1997)
Diakoloukas, V.D., Digalakis, V.V.: Adaptation of Hidden Markov Models using multiple stochastic transformations. In: Proc. European Conf. on Speech Communication and Technology, pp. 2063–2066 (1997)
Diakoloukas, V.D., Digalakis, V.V.: Maximum-likelihood stochastic-transformation adaptation of Hidden Markov-Models. IEEE Trans. Audio Speech Lang. Process. 7(2), 177–187 (1999)
Digalakis, V.V.: Online adaptation of hidden Markov models using incremental estimation algorithms. IEEE Trans. Audio Speech Lang. Process. 7(3), 253–261 (1999)
Eckert, W., Gallwitz, F., Niemann, H.: Combining stochastic and linguistic language models for recognition of spontaneous speech. In: Proc. Int. Conf. on Acoustics, Speech, and Signal Processing, Atlanta, vol. 1, pp. 423–426 (1996)
Fischer, A., Stahl, V.: Database and online adaptation for improved speech recognition in car environments. In: Proc. Int. Conf. on Acoustics, Speech, and Signal Processing, Phoenix, AZ (1999)
Gales, M.J.F., Woodland, P.C.: Variance compensation within the MLLR framework. Technical report, Cambridge University Engineering Department (1996)
Gauvain, J.-L., Lee, C.-H.: Maximum a posteriori estimation for multivariate Gaussian mixture observations of Markov chains. IEEE Trans. Audio Speech Lang. Process. 2(2), 291–298 (1994)
Huang, X., Acero, A., Hon, H.-W.: Spoken Language Processing: A Guide to Theory, Algorithm, and System Development. Prentice Hall, Englewood Cliffs (2001)
Iyer, R.M., Ostendorf, M.: Modeling long distance dependence in language: topic mixtures versus dynamic cache models. IEEE Trans. Audio Speech Lang. Process. 7(1), 30–39 (1999)
Kuhn, R., De Mori, R.: A cache-based natural language model for speech recognition. IEEE Trans. Pattern Anal. Mach. Intell. 12(6), 570–583 (1990)
Leggetter, C.J., Woodland, P.C.: Flexible speaker adaptation using maximum likelihood linear regression. In: Workshop on Spoken Language Systems Technology, pp. 110–115 (1995). ARPA
Leggetter, C.J., Woodland, P.C.: Maximum likelihood linear regression for speaker adaptation of continuous density Hidden Markov Models. Comput. Speech Lang. 9, 171–185 (1995)
Mahajan, M., Beeferman, D., Huang, X.D.: Improved topic-dependent language modeling using information retrieval techniques. In: Proc. Int. Conf. on Acoustics, Speech, and Signal Processing, Phoenix, AZ, vol. 1, pp. 15–19 (1999)
Martínez, F., Tapias, D., Álvarez, J.: Towards speech rate independence in large vocabulary continuous speech recognition. In: Proc. Int. Conf. on Acoustics, Speech, and Signal Processing, pp. 725–728 (1998)
Popovici, C., Baggia, P.: Specialized language models using dialogue predictions. In: Proc. Int. Conf. on Acoustics, Speech, and Signal Processing, München, vol. 2, pp. 779–782 (1997)
Siegler, M.A., Stern, R.M.: On the effects of speech rate in large vocabulary speech recognition systems. In: Proc. Int. Conf. on Acoustics, Speech, and Signal Processing, Detroit, vol. 1, pp. 612–615 (1995)
Wessel, F., Baader, A.: Robust dialogue-state dependent language modeling using leaving-one-out. In: Proc. Int. Conf. on Acoustics, Speech, and Signal Processing, Phoenix, AZ (1999)
Wessel, F., Baader, A., Ney, H.: A comparison of dialogue-state dependent language models. In: Proc. ECSA Workshop on Interactive Dialogue in Multi-Modal Systems, Irsee, Germany, pp. 93–96 (1999)
Wrede, B., Fink, G.A., Sagerer, G.: An investigation of modelling aspects for rate-dependent speech recognition. In: Proc. European Conf. on Speech Communication and Technology, Aalborg, vol. 4, pp. 2527–2530 (2001)
Author information
Authors and Affiliations
Rights and permissions
Copyright information
© 2014 Springer-Verlag London
About this chapter
Cite this chapter
Fink, G.A. (2014). Model Adaptation. In: Markov Models for Pattern Recognition. Advances in Computer Vision and Pattern Recognition. Springer, London. https://doi.org/10.1007/978-1-4471-6308-4_11
Download citation
DOI: https://doi.org/10.1007/978-1-4471-6308-4_11
Publisher Name: Springer, London
Print ISBN: 978-1-4471-6307-7
Online ISBN: 978-1-4471-6308-4
eBook Packages: Computer ScienceComputer Science (R0)