Soft state-tying for HMM-based speech recognition

This paper introduces a method for regularization of HMM sys-tems that avoids parameter overfitting causedby insufficient train-ing data. Regularization is done by augmenting the EM training method by a penalty term that favors simple and smooth HMM systems. The penalty term is constructed as a mixture model of negative exponential distributions that is assumed to generate the state dependent emission probabilities of the HMMs. This new method is the successful transfer of a well known regularization approachin neural networks to the HMM domain and can be inter-preted as a generalization of traditional state-tying for HMM sys-tems. The effect of regularization is demonstrated for continuous speech recognition tasks by improving overfitted triphone models and by speaker adaptation with limited training data. 1