Two-stage vector quantization-pyramidal lattice vector quantization and application to speech LSP coding

Motivated by two-stage vector quantization-(spherical) lattice vector quantization introduced by Pan and Fischer (1995), two-stage vector quantization-pyramidal lattice vector quantization (VQ-PLVQ) is proposed that has a lower computational requirement and a slightly superior encoding performance. The complexity of VQ-PLVQ is further reduced by introducing tree-structures into the first-stage VQ. It is found that the efficiency of tree-searching depends on source distributions. Tree-structured VQ-PLVQ provides a performance very close to VQ-PLVQ for sources with memory. Then, these two quantisation schemes are applied to encoding of speech line spectrum pair parameters. Tree-structured VQ-PLVQ is an attractive approach that achieves an outstanding performance for the spectral distortion of 1 dB as well as a significant reduction of the implementation complexity