In the paper an effective entropy coder designed for coding of prediction errors of audio signals is presented. The coder is implemented inside a greater structure, which signal modeling part is a lossless coding backward adaptation algorithm consisting of cascaded Ordinary Least Squares (OLS), three Normalized Least Mean Square (NLMS), and prediction error bias correction sections. The technique performance is compared to that of four other lossless codecs, including MPEG-4 Audio Lossless (ALS) one, and it is shown that indeed, on the average the new method is the best. The entropy coder is an advanced context adaptive Golomb one followed by two context adaptive arithmetic coders.
Novel ideas for lossless audio coding analyzed in the paper are linked with forward predictor adaptation, and concern optimization of predictors on the basis of zero-order entropy and MMAE criterions, and context sound coding. Direct use of the former criterion is linked with exponential growth of optimization procedure, hence, a suboptimal algorithm having polynomial complexity is proposed. It is shown that on average the new types of predictors are better than those obtained by MMSE technique, while two- and three context systems are on average better than a single predictor one. It also appears that 7-bit PARCOR coefficients in the MPEG-4 ALS standard have insufficient precision for some predictor length, and that for very long frames coding results improve with the predictor rank practically in unlimited way.
JavaScript jest wyłączony w Twojej przeglądarce internetowej. Włącz go, a następnie odśwież stronę, aby móc w pełni z niej korzystać.