A prediction performance between individual channels of a stereo signal is improved to improve a sound quality of a decoded signal. A first low pass filter LPF interrupts a high-range component of a first channel signal S1, and outputs a first low-range component S1′. A second low pass filter LPF interrupts a high-range component of a second channel signal S2, and outputs a second low-range component S2′. A predictor predicts the S2′ from the S1′, and outputs a prediction parameter composed of a delay time difference t and an amplitude ratio g. first channel encoder encodes the S1. A prediction parameter encoder encodes the prediction parameter. The encoded parameters of the encoded parameter of the S1 and the prediction parameter are then outputted.