To provide a voice synthesizing method that can obtain synthesized voice small in spectral distortion caused by the change of fundamental frequency.

Residual pitch waveform 302 generated from a reference voice signal 101, and an LCP coefficient 113 extracted from the reference voice signal 101 are stored. A voiced sound source signal 106 is generated in a voiced interval using the residual pitch waveform 108, and a voiceless sound source signal 105 is generated in a voiceless interval. A voice path filter 15 with the LPC coefficient 110 as a coefficient is driven by the voiced sound source signal 106 or the voiceless sound source signal 105 to generate a synthesized voice signal 109. In such a voice synthesizing method, a representative voice pitch 301 extracted from the reference voice signal 101 is filtered by an inverse filter part 33 according to the LPC coefficient 13 obtained by the LPC analysis of the reference voice signal 101 so as to generate the residual pitch waveform 302.