A method for audio encoding includes: analyzing an audio frame using a psychoacoustic model to obtain a corresponding masking curve and window information; transforming the audio frame according to the window information to obtain a spectrum, and dividing the spectrum into a plurality of frequency sub-bands; estimating a scale factor for each frequency sub-band; quantizing the frequency sub-bands; encoding the quantized frequency sub-bands; and packing the encoded frequency sub-bands and side information into an audio stream. Each scale factor is estimated based on a quantizable audio intensity of each frequency sub-band, which is adjusted according to a cumulative total amount of buffer space used for storing the encoded frequency sub-bands and an amount of buffer space used for storing a previously encoded audio frame, and a mean of intensities of all signals in the corresponding frequency sub-band and spectrum position of the corresponding frequency sub-band.