An estimation-theoretic framework is proposed for prediction in scalable coding of stereo-phonic audio, which overcomes fundamental drawbacks of conventional prediction approaches.The framework oers the means to combine all the information available at the enhancement-layer so as to produce the optimal signal estimate. The method e?ciently exploits both inter-channel and intra-channel redundancy without incurring additional side-information. Theproposed estimation-theoretic prediction is implemented within an MPEG-4 based scalablecoder for stereophonic audio signals. Objective and subjective performance comparison on theMPEG-4 SQAM database shows that optimal prediction yields substantial bit rate reductionwhile maintaining the same reproduction quality. The improvement is achieved at the cost ofa modest increase in computational complexity.
展开▼