An MPEG-4 Twin-VQ based high quality audio codec design

This paper presents a high quality audio codec design based on a transform-domain weighted interleave vector quantization (Twin-VQ) scheme adopted in the MPEG-4 audio standard. Three novel techniques are employed in this scheme to compress the data, ie, (1) flattening of MDCT coefficients by the spectrum of linear predictive coding (LPC) coefficients; (2) further flattening of MDCT coefficients by the Bark envelope; and (3) weighted interleave vector quantization. This paper examines the related design issues in implementing an efficient Twin-VQ codec. Fast computation algorithms are derived for the computationally intensive modules. Design parameters of each module are determined and the codebooks for weighted interleave vector quantization are constructed. Experimental results show that the designed codec can compress natural audio efficiently and reproduce high quality outputs.