Normalization of LPC residue for random access frame in audio coding

In a time-domain audio codec system, linear prediction is typically used as a tool to predict the value of future samples based on the value of previous samples. Such tool is useful to remove redundancy in the signal so that the signal can be more efficiently coded by an entropy coder. Modern digital audio system requires random access capability and because of this requirement, the residue produced by the predictor exhibits a high dynamic range. In this paper, an adaptive normalization scheme is proposed to reduce the energy level of first few prediction residues in random access blocks. This enhances the coding efficiency of the entropy coding. The proposed scheme can be used in speech/audio analysis and synthesis.