Perceptual Pre-weighting and Post-inverse weighting for Speech Coding

We investigate the effect on voice quality of perceptual pre-weighting of the input speech to a codec, and post- inverse weighting the output of the codec. The G.726 adaptive differential pulse code modulation (ADPCM) codec and the AMR narrowband (AMR-NB) code excited linear prediction (CELP) codec are employed in our experiments. The weighting function used has the same form as that of the perceptual weighting function for the analysis-by-synthesis codebook search in AMR- NB. We observe a significant improvement in voice quality at rates of 16 and 24 kbps in the case of G.726 when perceptual weighting is used. When we use pre-weighting with the AMR codec, the unweighted squared error is used within the analysis- by-synthesis codebook search loop, and we find that the quality of the pre-weighted approach is comparable to the quality achieved by the standard AMR codec. The proposed pre-weighting method requires an additional bit-rate of 1.35 kbps to communicate the linear prediction (LP) coefficients of the original speech input to the decoder.