This paper discusses two approaches for combining an efficient LR parser and phoneme-context-dependent HMM phone models and compares them through continuous speech recognition experiments. LR parsing is one of the most efficient parsing algorithms for speech recognition under grammatical constraints based on a context free grammar (CFG). To recognize continuous speech, it is advantageous to use allophone models for improving speech recognition accuracy, because they precisely represent allophonic variation caused by phoneme context. This paper aims to combine accurate allophonic models with the efficient LR parser to construct a powerful scheme for continuous speech recognition. In this paper, two phoneme-context-dependent LR parsing algorithms are proposed, which make it possible to drive allophonic HMMs. The algorithms are outlined as follows: (1) Algorithm for generating a phoneme-context-dependent LR parsing table using a CFG. (2) Algorithm for predicting the phoneme context dynamically in the LR parser by using a phoneme-context-independent LR table. This paper also includes discussion of the results of recognition experiments, and a comparison of performance and efficiency between these two algorithms. Keywords: continuous speech recognition, LR parser, phoneme environment, allophone, Hidden Markov Models, HMM-LR
Bibliographic reference. Nagai, Akito / Sagayama, Shigeki / Kita, Kenji (1991): "Phoneme-context-dependent LR parsing algorithms for HMM-based continuous speech recognition", In EUROSPEECH-1991, 1397-1400.