摘要 |
A filterbank (FA) breaks down the digital audio signal into a plurality of frequency subbands (SB), from which respective predictions are subtracted in an adder (SO1). The residues of the substraction are requantized in a quantizer (QU) under control of a bit allocator (BA) governed by a predetermined perceptual model in order to remove irrelevant information from the subband signals. The predictions are generated by circuit means (IQ, IS, CP, PR) based on the output signals from the quantizer, as a weighted average, by means of prediction coefficients (PCT), of a plurality of samples preceding in time by a delay interval (D) chosen for each subband within a time window. A formatter (FT) packs into frames the output signals of the quantizer together with the value of the delay interval. <IMAGE> |