Hui Li, Toshiyuki Takezawa, Harald Singer, Teruaki Hayashi
An Efficient Phoneme-Context-Dependent
LR Table Applied to the SSS-LR Continuous
Speech Recognition System
Abstract:In this report, we describe a method of generating a phoneme-context-dependent
canonical LR table through constraint propagation method (CPM) and its application to the SSS-LR continuous speech recognition system. For phrase recognition
tasks, a comparison of recognition performance for table level and parser level realization of GLR parsing algorithm is given. We also compare the performance of
SLR and CLR table for speech recognition.