摘要 |
PROBLEM TO BE SOLVED: To provide a technique for modeling a speech having desired rhythm characteristics. SOLUTION: The set of tags for defining rhythm characteristics is prepared, and the selected tag is arranged in the proper place of a text main body. Each tag imposes constraint on the rhythm characteristics of a speech to be generated by processing the text. The set of equations to be solved so that a curve for defining the rhythm characteristics across the range of words and phrases can be generated and the set of equations to be solved so that a curve for defining the rhythm characteristics of the respective words in the phrases can be generated are generated according to the processing of the speech text and the tag. The data defined by the curve can be used together with the text so that the speech having the rhythm characteristics defined by the tag can be generated. The set of tags is generated by the reading of the training text to be read by a target speaker, and a training corpus on which the rhythm characteristics of the target speaker are reflected is generated, and then the training corpus is analyzed so that the tag for modeling the rhythm characteristics of the training corpus can be generated. |