ISCA Archive Interspeech 2008
ISCA Archive Interspeech 2008

A shrinkage estimator for speech recognition with full covariance HMMs

Peter Bell, Simon King

We consider the problem of parameter estimation in full-covariance Gaussian mixture systems for automatic speech recognition. Due to the high dimensionality of the acoustic feature vector, the standard sample covariance matrix has a high variance and is often poorly-conditioned when the amount of training data is limited. We explain how the use of a shrinkage estimator can solve these problems, and derive a formula for the optimal shrinkage intensity. We present results of experiments on a phone recognition task, showing that the estimator gives a performance improvement over a standard full-covariance system.


doi: 10.21437/Interspeech.2008-106

Cite as: Bell, P., King, S. (2008) A shrinkage estimator for speech recognition with full covariance HMMs. Proc. Interspeech 2008, 910-913, doi: 10.21437/Interspeech.2008-106

@inproceedings{bell08_interspeech,
  author={Peter Bell and Simon King},
  title={{A shrinkage estimator for speech recognition with full covariance HMMs}},
  year=2008,
  booktitle={Proc. Interspeech 2008},
  pages={910--913},
  doi={10.21437/Interspeech.2008-106}
}