摘要 |
PROBLEM TO BE SOLVED: To change a pitch of an input speech waveform, while fully holding a characteristic of the input speech waveform. SOLUTION: A storage section 14 stores the input speech waveform WIN. A filter processing section 16 generates a processing object waveform WL by a low-pass filter processing for the input speech waveform WIN. A cycle specifying section 12 specifies a pitch cycle Ta of the input speech waveform WIN. A position detection section 32 sequentially detects a maximum amplitude position R for every pitch cycle Ta in the processing object waveform WL. A position correction section 34 determines a reference position Q by correcting a maximum amplitude position R so that variation of a phase by the low-pass filter processing may be compensated. A waveform extracting section 22 sequentially extracts a unit waveform U from the input speech waveform WIN based on a window function F in which a function value is a maximum value at the reference position Q. A waveform synthesis section 24 generates an output speech waveform WOUT by arranging each unit waveform U at intervals according to an output pitch POUT. COPYRIGHT: (C)2009,JPO&INPIT
|