Robust vector quantization based on spectral mapping with a noise estimate

Abstract

The problem of optimal quantization of a signal affected by additive noise in the general framework of vector quantization is studied. A noise estimator is used to adapt the vector quantization codebook to the specific noisy environment. Two adaptation strategies are discussed: switched adaptation (multicodebook) and an adaptation approach in which the codebook is computed from the optimal clean signals codebook and the noise estimate. Experimental results show a significant improvement over the direct vector quantization of the noisy signal.<>