In LPC analysis, a small number of coefficients, usually ten or twelve, are calculated from successive short segments of the speech wave. Since the articulators move slowly, the salient features of the speech wave also change slowly, and new sets of coefficients need only be calculated about once every 10 msec. Information about vocal excitation is also derived from the speech wave input. The LPC coefficients and excitation information are transmitted to the receiver, where they are used to predict new speech samples with minimum error. As one set of coefficients after another is received, the entire speech wave is resynthesised. The data rate needed to transmit the coefficients and the excitation information is much smaller than that required to transmit the speech waveform itself, resulting in transmission efficiencies.