To synthesize an arbitrary word which is almost natural by registering speech data in 1-syllable units and speech data spoken in daily-use KANJI (Chinese character) units and properly selecting and combining them together.

This system selects (n) syllables out of speech data generated by plural Japanese syllables, syllable by syllable, and plural syllables and registers speech data, spoken in (n)-syllable units (n: integer larger than 2) by all combinations, on a storage medium. For speech synthesis, the speech data are combined and plural speech data spoken with several kinds of intonation (speech pitch) for one kind of syllable are registered as speech data in 1-syllable units. For speech data in (n)-syllable units, plural speech data spoken while the intonation position is changed are registered for respective (n) syllables.