In this paper we apply a new method for quantizing the LPC parameters obtained from the analysis of different speech data. This new method uses the redundancy between the LPC coefficients of successive frames to get additional sub-quantization levels suitable for each coefficient group. This method is applied for english as well as for arabic sounds. The data under test comprises mono- and disyllabic sounds. The results show that the bit rate could be minimized in the case of maintaining the same error as that of the traditional method. If the bit rate is maintained the same, the error will be reduced.