Regressive linear prediction with doublet for speech signals

In this paper we proposed a new line prediction algorithm, Regressive Linear Prediction with Doublet (RLPD) for speech signals. Unlike conventional Linear Programming (LP), n
^th
sample of signal x(n) is predicted based on p + 2 previous samples by forming p regression lines. Each line is formed by using two alternative samples from p + 2 samples. As a result, the obtained all pole filter is able to perform better than conventional counterpart for small p. RLPD predictor is obtained by minimizing square of the prediction error. The experiment show that the performance of the RLPD method is effective compared to the LP method for the voiced speech signals corrupted with additive white gaussian noise.