Nonstationary linear prediction analysis of speech codec corrected by pre-stage forward volume normalizer

Linear prediction coding (LPC) technique for speech processing is the key part of speech codec. In this paper, forward volume normalizer (FVN) is proposed to keep the stationary characteristic of the speech signal fed to the encoder and to enhance the accuracy of LPC analysis under nonstationary speech condition. According to the measurement of ITU-T P.862 Perceptual Evaluation of Speech Quality (PESQ) equipment, the experimental results show the degradation of the subjective speech quality from nonstationary speech signal can be somewhat removed by the proposed Forward Volume Normalizer at the pre-stage of speech codec.