Linear Discriminant Analysis for Improved Large Vocabulary Continuous Speech Recognition

interaction of Linear Discriminant Analyand a modeling approach using continuous mixture density HMMs is studied experimentally. The largest improvements in speech recognition accuracy could be obtained when the classes for the LDA transform were defined to be sub-phone units. On a 12,000-word German recognition task with small overlap between training and test vocabulary a reduction in error rate by one fifth was achieved compared to the case without LDA. On the development set of the DARPA RM1 task the error rate wqs reduced by one third. For the DARPA speaker-dependent nogrammar case, the error rate averaged over 12 speakers was 9.9%. This was achieved with a recognizer employing LDA and a set of only 47 Viterbi-trained contextindependent phonemes.