An efficient embedded ADPCM coder

Embedded speech coding offers a graceful and acceptable degradation of speech quality when a network or channel becomes congested and transmitted information is lost. Unlike PCM which is inherently an embedded code, the more efficient differential codes such as DPCM and delta modulation (DM) do not possess embedded coding properties without modification. Embedded versions of both DPCM and DM have therefore been proposed [Goodman (1980), Wassell et al. (1988), Jayant (1983), Zhang and Lockhart (1991)]. The present authors demonstrate a novel embedded adaptive DPCM (EADPCM) speech coder. The coder uses the explicit noise coding technique [Jayant] to generate a supplemental bit stream which may be wholly or partly discarded during transmission. A new low-complexity algorithm is employed to match the supplemental adaptive PCM (APCM) coder to the ADPCM reconstruction error. The step size for the APCM coder is determined in two stages. First, a basic step size is estimated from the ADPCM step based on instantaneous or syllabic adaptation or their combination. This is followed by further instantaneous step size adjustment based on the APCM output. Simulation shows that a graceful reduction of reconstructed speech quality with progressive bit deletion is achieved over the range from 48 (or 40) kb/s to 16 kb/s in 8 kb/s steps.