Introduction to statistical pattern recognition (2nd ed.)
Introduction to statistical pattern recognition (2nd ed.)
Machine Learning
Tree-based state tying for high accuracy acoustic modelling
HLT '94 Proceedings of the workshop on Human Language Technology
| Hi-index | 0.00 |
A decision tree can be used not only as a classifier but also as a clustering method. One of such applications can be found in automatic speech recognition using hidden Markov models (HMMs). Due to the insufficient amount of training data, similar states of triphone HMMs are grouped together using a decision tree to share a common probability distribution. At the same time, in order to predict the statistics of unseen triphones, the decision tree is used as a classifier as well. In this paper, we study several cluster split criteria in decision tree building algorithms for the case where the instances to be clustered are probability density functions. Especially, when Gaussian probability distributions are to be clustered, we have found that the Bhattacharyya distance based measures are more consistent than the conventional log likelihood based measure.