A transform domain LMS audio coder

Abstract

In this paper, a transform domain LMS audio coder is presented where a transformed input is processed by a NLMS predictor. Three different transformations - Discrete Fourier transform (DFT), Discrete Cosine transform (DCT) and Discrete Wavelet transform (DWT) have been considered. The convergence performance of these predictors is observed with a synthesized music. The residuals of the predictors are coded and the coding efficiency of the Wavelet-based predictor is compared with predictors based on other transforms for synthesized music as well as for some real music signals. It is observed that audio coder with wavelet based predictor provides the least bit-rate in bits/sample. Besides this, the time complexity involved in prediction and coding using different transformations have been studied.