摘要 |
PROBLEM TO BE SOLVED: To create speech elements applicable to a speech synthesis device by rule with better accuracy and flexibility. SOLUTION: When a speech element is created by extracting a pitch waveform from a voiced sound speech waveform by a segmenting window in each pitch period, an interpolation characteristic on the spectrum of the segmenting window is set beforehand and the segmenting window is calculated according to the interpolation characteristic. Firstly, to the voiced sound speech waveform, a waveform local peak extraction part 1 marks a local peak of the waveform. Then a pitch period of the speech waveform is calculated. Next, the segmenting window is calculated by a segmenting window calculation part 2. As a setting of an interpolation characteristic on the spectrum of the segmenting window, the segmenting window is calculated by, for example, an inverse discrete Fourier transform. Finally, a pitch waveform extraction part 3 extracts the pitch waveform by applying the window calculated by the segmenting window calculation part centering the local peak in each pitch period of the speech waveform. |