A candidate voice segment sequence generator 1 generates candidate voice segment sequences 102 for an input language information sequence 101 by using DB voice segments 105 in a voice segment database 4. An output voice segment sequence determinator 2 calculates a degree of match between the input language information sequence 101 and each of the candidate voice segment sequences 102 by using a parameter 107 showing a value according to a cooccurrence criterion 106 for cooccurrence between the input language information sequence 101 and a sound parameter showing the attribute of each of a plurality of candidate voice segments in each of the candidate voice segment sequences 102, and determines an output voice segment sequence 103 on the basis of the degree of match.