Audio Signal Processing and Coding

7.7 DIFFERENTIAL PERCEPTUAL AUDIO CODER

Other investigators have also developed promising schemes for transform coding of audio. Paraskevas and Mourjopoulos [Para95] reported on a differential perceptual audio coder (DPAC), which makes use of a novel scheme for exploiting long-term correlations. DPAC works as follows. Input audio is transformed using the MDCT. A two-state classifier then labels each new frame of transform coefficients as either a “reference” frame or a “simple” frame. The classifier labels as “reference” frames that contain significant audible differences from the previous frame. The classifier labels nonreference frames as “simple.” Reference frames are quantized and encoded using scalar quantization and psychoacoustic bit allocation strategies similar to Johnston's PXFM. Simple frames, however, are subjected to coefficient substitution. Coefficients whose magnitude differences with respect to the previous reference frame are below an experimentally optimized threshold are replaced at the decoder by the corresponding reference frame coefficients. The encoder, then, replaces subthreshold coefficients with zeros, thus saving transmission bits. Unlike the interframe predictive coding schemes of Mahieux and Petit, the DPAC coefficient substitution system is advantageous in that it guarantees that the “simple” frame bit allocation will always be less than or equal to the bit allocation that would be required if the frame was coded as a “reference” frame. Suprathreshold ...