PURPOSE: To provide a unit voice selection system wherein a more natural, easy-to-hear synthesized voice is created by using continuous time length information and basic frequency information when a unit voice is selected.

CONSTITUTION: A unit parameter selecting section 6 evaluates a coincidence status, in comparison, between a phoneme line before and after a unit voice in an input character line and a similar phoneme line in a unit voice data for each of unit voice data in a given unit voice data table 3. At the same time, continuous time length information accumulated in a continuous time length information accumulating section 11 about a continuous time prescribed unit in a unit voice is compared with a continuous time determined by a continuous time length determining section 9 in a continuous time length unit prescribed unit for evaluation, and a unit voice which has the highest total evaluation value in combination of both evaluation values is selected as an optimum unit voice parameter. The selected unit voice parameter is outputted to a voice synthesizing section 7.