Third European Conference on Speech Communication and Technology

Berlin, Germany
September 22-25, 1993


KARS: A Speaker-Independent, Vocabulary-Independent Speech Recognition System

Myoung-Wan Koo

Software Research Labs, Korea Telecom, Seoul, Korea

In this paper, we present an information retrieval system based on recognition technology (KARS: Korea Telecom Automatic Recognition System). If a user commands isolated Korean words over the telephone according to the voice instruction, the system announces telephone numbers, locations and activities of sections in Korea Telecom Research Center. The system is a HMM (hidden Markov model)-based speaker-independent, vocabulary-independent isolated speech recognizer with 129 word recognition vocabularies. For the vocabulary-independence, we use the condext-dependent phones and also present the feature extraction method in which the features for training may be different from those for recognition. We have achieved speaker-independent, vocabulary-independent recognition results of 92.6%.

Keywords: Vocabulary-Independence, Speech Recognition System, EMM

Full Paper

Bibliographic reference.  Koo, Myoung-Wan (1993): "KARS: a speaker-independent, vocabulary-independent speech recognition system", In EUROSPEECH'93, 1837-1840.