Audio Signal Processing and Coding

7.6 ADAPTIVE SPECTRAL ENTROPY CODING

The MSC, OCF, PXFM, Brandenburg-Johnston hybrid, and CNET transform coders were eventually clustered into a single proposal by the ISO/IEC JTC1/SC2 WG11 committee. As a result, Schroeder, Brandenburg, Johnston, Herre, and Mahieux collaborated in 1991 to propose for acceptance as the new MPEG audio compression standard a flexible coding algorithm, ASPEC, which incorporated the best features of each coder in the group. ASPEC [Bran91] was claimed to produce better quality than any of the individual coders at 64 kb/s.

The structure of ASPEC combines elements from all of its predecessors. Like OCF and the later CNET coders, ASPEC uses the MDCT for time-frequency mapping. The masking model is similar to that used in PXFM and the Brandenburg-Johnston hybrid coders, including the sophisticated tonality estimation scheme at lower bit rates. The quantization and coding procedures use the pair of nested loops proposed for OCF, as well as the block differential coding scheme developed at CNET. Moreover, long runs of masked coefficients are run-length and Huffman encoded. Quantized scale factors and transform coefficients are Huffman coded also. Pre-echoes are controlled using a dynamic window switching mechanism, like the Thomson coder [Edle89]. ASPEC offers several modes for different quality levels, ranging from 64 to 192 kb/s per channel. A real-time ASPEC implementation for coding one channel at 64 kb/s was realized on a pair of 33-MHz Motorola DSP56001 ...