A method for providing synthesized speech using regular-pulse excitation includes a first step (300) of processing input speech to provide a residual excitation signal. A next step (302) includes defining important samples of the residual signal. Low frequency residual signals are particularly important. A next step (304) includes coding the important samples using regular-pulse excitation. A next step includes storing the important samples to random regular-pulse excitation grid positions in a memory using a first set of pseudorandomly generated numbers to assign the grid positions of each of the important samples. In this way, code rate for controlled, voice-only signals can be increased. This best applies to non-real time speech storage of voice tags, prompts and messaging.