EFFICIENT BACKWARD CODING FOR WIDE-BAND SPEECH AND MUSIC SIGNALS (ADPCM-AB).

This paper proposes a new backward speech coding system (ADPCM-AB) for wideband (about 8 kHz) speech and music signals. This system uses a split-band adaptive predictive coding scheme with a PARCOR lattice filter and backward dynamic bit allocation over subbands and stereo channels. Linear prediction coefficients and dynamic bit allocation parameters are calculated using locally decoded residual signals. Accordingly, delay time is only 5 milliseconds and transmission of side-information parameters is unnecessary. Paired comparison listening tests confirm that the speech quality of ADPCM-AB is very close to that of forward coding system.