VECTOR EXCITATION SPEECH OR AUDIO CODER FOR TRANSMISSION OR STORAGE

摘要

A vector excitation coder efficiently com-presses vectors by using an optimum codebook designedoff line, using an initial arbitrary codebook and aset of speech training vectors exploiting codevectorsparsity (i.e., by making zero all but a selectednumber of samples of lowest amplitude in each of Ncodebook vectors). A fast-search method selects anumber Nc of good excitation vectors from the code-book, where Nc is much smaller than N, and uses onlythe Nc vectors in an exhaustive search for the bestmatch between a perceptually weighted input vectorZn~ and an estimate Zn derived from a codebook vectorprocessed through long-term and short-terms filters,and a perceptual weighting filter. The zero inputresponse of these cascaded filters is calculated andsubtracted from an input speech vector sn after per-ceptual weighting to produce a vector rn. The code-book search operation is performed using ,by calculating the numerator of a fast inner productand calculating the denominator by a fast inner prod-uct for each codebook vector cj, computing the righthand side of the equation once per frame, and thencross multiplying the numerators and denominators todetermine if N2/D2 is less than N1/D1 by determiningif N1D2N2D1. If not N2 and D2 replace N1 and D1 inregisters En and Ed.