PURPOSE: To save the labor for manually applying the local peak position, allow the real-time process, prevent the increase of the data capacity for providing the local peak position in advance, and provide the high-speed voice pitch conversion process by automatically applying the local peak position to the voice wave-form based on power.

CONSTITUTION: The short-segment power calculated via the frame length shorter than the pitch period of the input voice and the threshold value power calculated via the frame length several times the pitch period are obtained synchronously with the time axis of the input voice by a short-segment power calculation section 13 and a threshold value power calculation section 14, the segment where the short-segment power is larger than the threshold value power is obtained as the local peak position search segment on the voice wave-form by a peak position search segment extraction section 15, and the maximum or minimum point where the magnitude of the amplitude on the voice wave-form becomes the largest in the search segment is determined as the local peak position by a voice wave-form peak position extracting section 16.