Skip to Main Content
In the paper, we describe a lattice vector quantization model based on scalable and high order codebook extension scheme. In the scheme, we expand the basic codebook so that the expanded codebook contains the point to be quantized, and then fractionize the expanded Voronoi cell, and use the subdivided lattice point to quantize the point, thereby reducing increasing quantization distortion caused by cell-amplification. By repeating such a 'cell extension - cell subdivision - vector quantization' process, we can gradually increase the quantization precision of outside points and achieve progressively refined quantization at each stage of codebook extension. We demonstrate the efficiency of the proposed scheme in the application of speech spectrum quantization. Experiments on encoding multiple low bitrates audio sequences show that the scheme provides superior performance with moderate increase in complexity when compared with conventional lattice vector quantization by Voronoi extension used in AMR-WB+ audio codec.