To provide a method, device, etc., for audio encoding.

An audio signal (x) to be encoded is received and converted from a time range to a frequency range, and an audio signal having been quantized is generated from the signal (x) having been converted together with a set of long-period prediction coefficients A, and the current time frame of the received signal (x) is predicted directly from the last time frame or a frame before it by using a coefficient A. The predicted audio signal generated by using the coefficient A is converted from a time range to a frequency range and each frequency subband is compared with that of the signal (x) for generating an error signal E(k). The signal E(k) is quantized to generate a set of error signals having been quantized and combined with the coefficient A for generating an audio signal having been encoded.