Provided is a method of encoding an audio/speech signal, the method including determining a variable length of a frame, that is, a processing unit of an input signal in accordance with a position of an attack in the input signal; transforming each frame of the input signal to a frequency domain and dividing the frame into a plurality of sub frequency bands; and, if a signal of a sub frequency band is determined to be encoded in the frequency domain, encoding the signal of the sub frequency band in the frequency domain, and if the signal of the sub frequency band is determined to be encoded in a time domain, inverse transforming the signal of the sub frequency band to the time domain and encoding the inverse transformed signal in the time domain. According to the present invention, the audio/speech signal may be efficiently encoded by controlling time resolution and frequency resolution.