Forward and Backward Design of Predictors for Lossless Audio Coding

In this paper we propose a cascade prediction scheme to be used in predictive lossless coding. First a linear predictor is designed using Levinson-Durbin algorithm, using a frame of data which is not available to the decoder, the predictor coefficients being sent as side information. The second predictor is an and adaptive, context based L predictor, designed using only already transmitted data. The DPCM-like coding scheme is completed with a context based GolombRice coder. The performance of the encoding scheme for lossless audio compression is experimentally evaluated for several standard speech and audio files.