Fast search techniques for use in vector quantisation and related pattern matching applications

Two techniques for the design and fast search of a vector quantiser codebook are proposed. These classify speech data vectors according to (i) the sign and (ii) the slope between successive samples. With both techniques the performance obtained is superior to that obtained from a conventional gain/shape vector quantiser and comparable to that of a full search vector quantiser. A major attraction of both approaches is that they significantly reduce the number of computations required for the codebook search.