Transform coding at 4.8 kbit/sec using interleaving of transform frames and dual gain-shape vector quantisation

The dual gain shape VQ (vector quantization) transform coding algorithm refines the distribution of quantization resources by supplementing interband bit allocation with intraband reallocation, and, by introducing two levels of vector normalization, allows the design of more efficient codebooks. Better use of limited bit resources at the coding rate of 4.8 kbit/s is thus achieved, yielding a 1 dB improvement in coding performance. The introduction of frame overlapping in the transform domain, by implementing a continuous frame interleave, gives significant improvements in the subjective quality of the coded speech. The effect produced by interleaving could be explained as an exploitation of interframe correlation, in creating vectors for VQ, and as an implicit promotion of frame interdependency in bit allocations.<>