PURPOSE:To encode a voice with small waveform distortion under conditions of a small information amount and a short delay time by approximating a signal of a current frame with cycles obtained by analyzing a past decoded waveform. CONSTITUTION:In a method which encodes a voice in units of frames each consisting of a constant number of samples, a linear prediction coefficient which is found by linear prediction is set in a composing filter and such a sequence of a residue driving sound source signal that a sequence of a decoded signal obtained by passing the residue driving sound source signal through the composing filter is closest to the voice in a current frame is searched for in a sequence of past decoded residue signals, i.e. residue signals with cycle values tau0, tau1,... tauN-1. At this time, M candidates for cycle values are selected in, for example, decreasing order of autocorrelation by analyzing the correlation of the residue signals to generate a correspondence table of the cycle values and intermediate codes, and the cycle value which is closest to the voice in the current frame is selected among the M candidates. The cycle value with high correlation is selected, so the information amount can be reduced almost to half.
展开▼
