摘要 |
A plurality of sound pitch period transition channels are extracted by means of a sound pitch tracker (11) over a sequence, and a channel with a minimum mean prediction gain over the sequence is chosen from among the extracted channels. A subsequent preliminary sound pitch selection can be made during subsequence processing so as to select a plurality of candidates close to the sound pitch of the selected transition channel for each subsystem by using the internal product of the input speech signal and each code vector (21, 22). Finally, a sound pitch period having minimum waveform distortion is chosen (28) for each subsequence. <IMAGE>
|