The primary problem in large vocabulary conversational speech recognition (LVCSR) is poor acoustic-level matching due to large variability in pronunciations. There is much to explore about the "quality" of states in an HMM and the inter-relationships between inter-state and intra-state Gaussians used to model speech. Of particular interest is the variable discriminating power of the individual states. The fundamental concept addressed in this paper is to investigate means of exploiting such dependencies through model topology optimization based on the Bayesian Information Criterion (BIC) and the Minimum Description Length (MDL) principle.
Cite as: Hamaker, J., Ganapathiraju, A., Picone, J. (1998) Information theoretic approaches to model selection. Proc. 5th International Conference on Spoken Language Processing (ICSLP 1998), paper 0653, doi: 10.21437/ICSLP.1998-178
@inproceedings{hamaker98_icslp,
author={Jonathan Hamaker and Aravind Ganapathiraju and Joseph Picone},
title={{Information theoretic approaches to model selection}},
year=1998,
booktitle={Proc. 5th International Conference on Spoken Language Processing (ICSLP 1998)},
pages={paper 0653},
doi={10.21437/ICSLP.1998-178}
}