According to the invention, an excitation signal (EXC) is generated as a result of sampled excitation values in order to excite an audio synthesis filter (ASYN), the generated sampled excitation values being continuously stored in an adaptive codebook (ACB). A noise generator (NOISE) is provided which continuously generates random sampled values. A sequence (EXC_P) of the stored sampled excitation values is selected from the adaptive codebook (ACB) based on a fed audio fundamental frequency parameter (PITCH) by means of which a time gap between the sequence (EXC_P) that is to be selected and the actual time reference is predefined. The excitation signal (EXC) is generated by mixing the selected sequence (EXC_P) with a random sequence (EXC_N) encompassing actual random sampled valued of the noise generator.