A New Perceptual and Adaptive Quantization Method for Line Spectral Frequency Parameters Based on Noise-to-Mask Ratio

摘要

In this paper, a new quantization method for line spectral frequency (LSF) parameters based on the perceptually weighted distance measure is presented. We derive the proosed distance measure from the definition of average noise-to-mask ratio (NMR) and show that it can be reduced to the form of weighted LSF distance measure having perceptually meaningful weights. We also propose an adaptive quantization scheme for LSF parameters. In our scheme, the bit rates are minimized by choosing the bit configuration with the lowest total bits whose maximum NMR has negative value. By jointly combining the proposed distance measure and the adaptive bit allocation scheme, the proposed system requires only 862 bps for obtaining 94