Language model cross adaptation for LVCSR system combination
Computer Speech and Language
Hi-index | 0.00 |
This paper describes recent changes in Dragon's speech recognition system which have markedly improved performance on conversational telephone speech. Key changes include: the conversion to modified perceptual linear prediction (PLP)-based cepstra from mel-cepstra; the replacement of our usual IMELDA transformation by a new transform using "semi-tied covariance"; a new multi-pass adaptation protocol; probabilities on alternate pronunciations in the lexicon; the addition of word-boundary tags in our acoustic models and the redistribution of model parameters to build fewer output distributions but with more mixture components per model.