| 摘要 |
<p>PURPOSE: To reduce a maximum point detecting process for a part which overlaps in a pitch extraction section and shorten the processing time for pitch extraction by finding a maximum point in a specified small section and temporarily storing the maximum point in the small section in a memory, and selecting the largest value from among maximum points. CONSTITUTION: Assuming that a maximum point detection section Ta is divided into J parts and a pitch extraction section Ti is divided into K small sections, a 1st maximum point detecting means 1 finds maximum points Mj by the J small sections from an inputted speech waveform, and the maximum points are stored in a storage means which temporarily stores the maximum points of all the small sections of the constant section Ta by shifting them each time a found maximum point of a small section is inputted. Then a 2nd maximum point detecting means 3 finds the maximum point M in the maximum point detection section Ta from among the J maximum points Mj. Namely, while a speech waveform is stored in a memory, a maximum point M is found through another route so as to find an autocorrelation function. Consequently, the memory is not used to detect the maximum point M and the memory is dedicated to pitch extraction, so the pitch extraction time is shortened.</p> |
