H90-1067 for comparison . In both cases , cooccurrence smoothing was performed along with deleted
H90-1060 in the training . However , the cooccurrence smoothing is not appropriate for models
P94-1038 differences between our method and cooccurrence smoothing . A more detailed analysis would
H89-1010 performed with and without triphone cooccurrence smoothing on all three test sets . These
P97-1008 improvement in error rate over cooccurrence smoothing . We also investigated the effect
E97-1008 improvement in error rate over cooccurrence smoothing . We also investigated the effect
P94-1038 Sterling ( 1993 ) applied the cooccurrence smoothing technique to estimate the likelihood
H90-1060 discrete densities , triphone cooccurrence smoothing \ -LSB- 7 \ -RSB- , has proven
P94-1038 the same corpus and task . In cooccurrence smoothing , as in our method , a baseline
P94-1038 for all bigrams to combine the cooccurrence smoothing model with MLE models of bigrams
H90-1060 matrices . As such , 11Sphone cooccurrence smoothing is a speaker-specific modeling
P97-1008 Pereira , and Lee ( 1994 ) and the cooccurrence smoothing method of Essen and Steinbiss
E97-1008 Pereira , and Lee ( 1994 ) and the cooccurrence smoothing method of Essen and Steinbiss
P94-1038 sufficiently similar words , whereas the cooccurrence smoothing method sums over all words in
H90-1060 modeling techniques such as triphone cooccurrence smoothing and probabilistic spectral transfonnation
P94-1038 estimates . The similarity model in cooccurrence smoothing is based on the intuition that
H89-1010 cooccurrence smoothing , and triphone cooccurrence smoothing . These methods were presented
P94-1038 written text . Related Work The cooccurrence smoothing technique ( Essen and Steinbiss
P94-1038 used for language modeling in the cooccurrence smoothing method of Essen and Steinbiss
P94-1038 of Centrust would have reported cooccurrence smoothing is given by Ps ( wzIwi ) = EP
hide detail