摘要 |
<P>PROBLEM TO BE SOLVED: To provide a speech processing apparatus and method, and a program, capable of generating a natural pitch contour which smoothly changes. <P>SOLUTION: Based on time length for each character string in each linguistic level which is included in an input document, a basic frequency of speech corresponding to the input document is divided into a plurality of segments, linear transformation of a segment group for each linguistic level is performed by a predetermined operator in which inverse transformation is possible, and a first parameter group according to each linguistic level is generated. Moreover, for each character string in each linguistic level included in the input document, a descriptor which shows features of the character string is generated, and the first parameter in each of the linguistic level is clustered based on the descriptor corresponding to the linguistic level, and model learning is performed as pitch contour model for each linguistic level. <P>COPYRIGHT: (C)2010,JPO&INPIT |