A large size Korean speech database under construction at ETRI is introduced. We have three kinds of speech databases. They are 35 connected-4-digits,144 CV monosyllables which are segmented into phonemes(C+V)combination, and 445 phoneme-balanced words. The first two are collected from 10 male speakers and the last, from 4 male and 4 female speakers. For the easy application of the database in speech research, we proposed a 2-level acoustic-phonetic transcription to express almost all possible phonetic environments in Korean. And the transcriptions were carried out manually. For the effective management of these databases, a relational database management system, which helps fast access and easy manipulation, was used.