Second International Conference on Spoken Language Processing (ICSLP'92)

Banff, Alberta, Canada
October 13-16, 1992

A Frame-Synchronous Continuous Speech Recognition Algorithm Using a Top-Down Parsing of Context-Free Grammar

Atsuhiko Kai, Seiichi Nakagawa

Department of Information and Computer Sciences, Toyohashi University of Technology, Toyohashi, Japan

In this paper, a frame-synchronous continuous speech recognition algorithm using a context-free grammar and its evaluation are described. A frame-synchronous parsing algorithm for context-free grammar based on topdown strategy is used for the prediction of successive words. The parsing process is incorporated into the One-Pass search algorithm [1] with the use of finite-state-automaton. Of course, since the number of the states becomes large, we should use the beam search method on the prediction and prune unreliable candidate trees in the search space. In the evaluation experiments, we used syllable-based continuous density HMMs. By using the proposed algorithm, we obtained the sentence recognition rate of 90% in a speaker adaptation mode, while our conventional system based on word spotting and lattice parsing algorithm obtained 80.7% for the task of perplexity 10.

Full Paper

Bibliographic reference.  Kai, Atsuhiko / Nakagawa, Seiichi (1992): "A frame-synchronous continuous speech recognition algorithm using a top-down parsing of context-free grammar", In ICSLP-1992, 257-260.