Lossless speech coding based on ITU-T G.729 standard using binary discrete cosine transform (BinDCT) and integer adaptive filtering (IntAF).

摘要

Lossless coding is a topic of high interest for researchers as well as for commercial applications. A multi-layered scheme based on ITU-T G.729 standard, comprising of a base layer and a lossless enhancement layer, is proposed here for lossless speech coding. The G.729 bit-stream forms the base layer, and the residual signal extracted from G.729 encoder forms the lossless enhancement layer. Three different schemes are proposed for the lossless compression of the residual signal. The first scheme implements a modified form of Binary Discrete Cosine Transform (BinDCT), the second scheme uses a novel Integer Adaptive Filtering (IntAF) using subband decomposition, and the third scheme uses BinDCT and IntAF in cascade. The proposed schemes have Integer-to-Integer mapping property and provide perfect reconstruction (PR). The output of each of these schemes is losslessly entropy coded using Huffman coding. A detailed coding gain analysis and computational complexity analysis is presented for each of the proposed schemes. The compression ratios are obtained for various test speech samples by computer simulation of the proposed schemes. Finally, a comparison of the compression ratios obtained using other lossless speech codecs and that obtained using the proposed schemes is presented.