This paper describes a newly developed Japanese text-to-speech conversion system. The phonetic and prosodic information for speech synthesis is extracted from the input text by the Japanese text analyzer so that Kanji and Kana texts can be input directly to the system. The synthetic speech is generated by concatenating speech synthesis units which contain symmetrical waveforms. As synthesis units, VCV syllables are used as well as CV syllables and symmetrical waveforms have been obtained by PSE analysis. Therefore, the synthetic speech sounds smooth and natural. The system has been built on a compact board which can be housed in a PC slot.