Near-optimum linear predictive speech coding

A near-optimum linear-predictive speech coding scheme is proposed. Near-optimum performance is defined in terms of minimizing a perceptually weighted mean-squared-error distortion measured directly between the original speech and the reconstructed speech. This is achieved by an efficient method which involves iteratively applying closed-loop analysis methods (or analysis-by-analysis methods) to the spectrum filter and the excitation signal. It is shown that this method can be applied to any linear predictive coder (LPC) with any perceptual distortion measure and it is especially useful for low-rate speech coders (less than 4.8 kb/s) where analysis-by-synthesis techniques become essential