To prevent a voice section in an input voice signal from being misdetected due to unsteady noise.

By this voice section detecting method which decides whether or not a current frame is a voice section by analyzing the input signal by frames, a threshold (Th) for voice detection is found from reference power (SP) and a deviation (Var) based upon the difference (Std) between the maximum value of the absolute values of power of all samples present in the current frame and the reference power (SP), and it is decided whether or not the current frame is a voice section by comparing the absolute values of amplitudes of all the samples present in the current frame or the maximum value of power of all the samples with the threshold (Th) for voice detection, thereby detecting the voice section C.