Frame synthesis for speech signal in code excited linear predictor

The method involves starting with an ideal regular pulse excitation sequence from which parameters are taken to form the stochastic components of the excitation vectors for a synthesis filter (12), yielding the reconstructed target signal. The parameters include the position of the first non-zero pulse in the sequence, the positions of a preselected number of largest pulses, and their amplitudes and signs. The number the of largest pulses is selected to be from one-sixth to one-quarter of the number of sample values in a subframe of the analytical frame. The stochastic excitation components are computed afresh for each subframe.