Frequency-space contractive speech coding

A very efficient speech coding scheme with a compression ratio of up to 120:1 is described. The difference between the fast Fourier transform (FFT) of two consecutive frames of speech is considered to be a two-dimensional image in the frequency domain. Using linear contractive transformation; this image can be reduced to a set of rules and coefficients that could either be transmitted or stored. The speech signal is reconstructed using these coefficients with minimal loss of signal quality. The output speech is superior in quality to that of linear predictive coding (LPC) speech. This algorithm can be applied equally to one-and two-dimensional signals, providing a possible solution for the hypermedia problem