On the predictor for the waveform coding of speech signals by using the dual first order difference values

It is argued that in the autocorrelation of speech samples, the autocorrelation of the adjacent past and the next sample is larger than the autocorrelation of several-order time-delayed samples. Thus, it is more effective to use the adjacent past and the next sample for prediction of the present sample, rather than use only the several-order time-delayed past. A new predictor that predicts the present sample by using the one past and the next sample is proposed.<
>