This paper describes a preliminary version of the text-to-speech system for Korean, which is referred to as "Geul-Sori". Input letters include Korean characters, numerals, and punctation marks. As a synthetic unit, we use the demisyllables and the control parameters of prosody are generated by rule. Geul-Sori was implemented on IBM-PC/AT using the TMS320C25 DSP chip. Geul-Sori is well operated, but yet poor in naturalness of speech. We are now developing many kinds of rules to achieve more naturalness and more intelligibility of the synthesized speech.