Linear contractivity speech coding

Abstract

The authors describe an elegant coding scheme that resulted in more than a 120:1 compression ratio of the speech signal. This technique is derived from the linear contractivity method of image coding and generation. The difference between the fast Fourier transforms (FFTs) of two consecutive speech frames is considered to be a two-dimensional image, then through various transformations a set of rules that represent this image are deduced. These rules can be used to reconstruct this difference and ultimately the speech signal itself with minimum loss in the signal content.<>