An experimental field trial of a large vocabulary speaker independent recognition system

Abstract

Presents a large vocabulary, speaker independent speech recognition system (KT-STOCK) and describes its experimental field trial over the telephone network. KT-STOCK is an information retrieval system with which one can obtain the current price of a stock by saying a stock name among 712 stock names listed on the Korea stock exchange. The system is a HMM (hidden Markov model)-based isolated speech recognizer which uses the phoneme-like unit as a basic unit. KT-STOCK has been put under experimental field trials since June 24 1994. Preliminary results show that the performance in the simulation environment is different from that in the real environment. Currently, the authors have achieved the recognition rate of 61.9% in the real environment.<>