Singular value excitation speech coder

Abstract

A new low delay speech coder is proposed. In order to achieve low delay, the coder is based on the memory dependent vector quantization of the synthesis filter. Another basic principle of the proposed scheme is to transform the excitation in a domain where the uncorrelated excitation samples are ordered according to their decreasing level of performance; this is achieved by means of a singular value decomposition transformation. The transformed samples are considered to be a sequence of Laplacian random variables that are vector quantized in an efficient manner using a geometrical lattice VQ. The coder was tested at a bit-rate of 8625 bit/s and a delay of 8 ms; a good reproduced voice quality was obtained.