摘要 |
PROBLEM TO BE SOLVED: To enable a voice synthesizer to be applied to the line of all phonemes regardless of the line of a sound voice phoneme and an unvoiced phoneme although the device is a device having a small number of unit patterns. SOLUTION: A pitch pattern to be obtained by analyzing the voice uttered by a human being is approximated to a pitch pattern by a model based on a prescribed function and a unit pattern shape in which the maximum value of pitch frequencies of unit patterns cut out in a prescribed unit from the approximated pitch pattern which is generated from the model parameter used in the approximation is made to be a reference is plural stored in a storage part 109. Then, a unit pattern shape is selected from the storage part 109 by a selecting and processing part 110 based on the analyzed result of a text analyzing part 101 with respect to an input text by using a prescribed rule and also a pitch bias amount is estimated by an estimating and processing part 111 by using a prescribed rule and unit patterns are generated by a pitch pattern generating part 108 from this unit pattern shape and the pitch bias amount and then a pitch pattern is generated from the unit patterns in the part 108. |