Disclosed is a voice decoding apparatus wherein the processor may be continuously employed for other applications for a prescribed time but, in response to an urgent interrupt, the processor can generate synthesised sound even when being used for other applications, without interruption. In this apparatus, a packet receiving section (101) receives packets of the layers of a plurality of frames and extracts code from the received packets. A state/code storage section (103) stores the code and decoding state of the code. A layer selection section (104) selects a layer number and a frame number corresponding to the code to be initially decoded, based on the decoding state. A decoding section (105) decodes the code of the selected frame number and layer number.