H89-1016 effectiveness of our extension of the corrective training algorithm to speaker-independent
H89-2039 and recognition 4 . Some form of corrective training to improve word discrimination
H89-1016 rate by 24-44 % . We modified the corrective training algorithm \ -LSB- 1 \ -RSB- for
H90-1060 model for confusable words . No corrective training was used for the BYBLOS results
H90-1063 have only applied our discrete corrective training algorithm \ -LSB- 15 \ -RSB-
H89-1016 that our extension of the IBM corrective training algorithm to continuous speech
H90-1063 improvement . Finally , we investigated corrective training for semicontinuous models . At
H90-1039 model . In some forms , including corrective training \ -LSB- 2 \ -RSB- , it is performed
H89-2008 phonological rules , and a new corrective training procedure . Lexical Expansion
H90-1039 problem . Evidence showing that corrective training inserts the training language
H89-1016 while significant , suggests that corrective training becomes overly specialized for
H89-1016 continuous speech recognition . Corrective training reduced SPHINX 's error rate
H89-1016 -LSB- 1 \ -RSB- introduced the corrective training algorithm for HMMs as an alternative
H90-1063 considers top N codewords , while the corrective training uses only the top 1 codeword
H89-2039 is primarily due to the use of corrective training and inter-word units . OVERALL
H89-2037 triphone models \ -LSB- 9 \ -RSB- and corrective training \ -LSB- 10 \ -RSB- were not used
H89-1024 problem and become , in effect , corrective training \ -LSB- 12 \ -RSB- on the test
H89-2032 modeling , interword CD PLUs , corrective training , and multiple lexical entry
H90-1058 sex-specific recognition models and from corrective training . SSI explored tree-structured
H89-1016 The third and final stage uses corrective training to refine the dis cnminatory
hide detail