A Korean Large Vocabulary Speech Recognition System for Automatic Telephone Number Query Service

In this paper, we introduce a Korean large vocabulary speech recognition system which can recognize sentence utterances with vocabulary size of 1160 words, and be used for automatic telephone number query service. This system consists of four sub-systems. The first one is an acoustic processor recognizing words in an input sentence by Hidden Markov Mode(HMM) based speech recognition algorithm. The second one is a linguistic processor which estimates input sentence from the result of the acoustic processor and determines next words after a word by using the syntactic information. The third one is a time reduction processor reducing recognition time by limiting the number of candidate words to be computed in the acoustic processor. The time reduction processor uses linguistic information and acoustic information contained in an input sentence. The last one is a speaker adaptation processor which adapts parameters of the speech recognition system to new speakers as soon as possible. The last subsystem uses VQ adaptation and HMM parameter adaptation based on the spectral mapping. We also present out recent works to improve the performance of the large vocabulary speech recognition system. These works are focused on the enhancement of the acoustic processor and the time reduction processor for speaker-independent speech recognition. New approach for the speaker adaptation is also described.