In an auto‐correlation vocoder system, samples of the short‐time auto‐correlation function of a speech signal are transmitted over a reduced bandwidth channel and converted into a time wave that is properly periodic and a replica of the auto‐correlation function. By Wiener's theorem, the amplitude spectrum of the thus synthesized signal is the power spectrum of the analyzed speech signal. Phenomenologically, this spectrum squaring has four distinct effects: 1. Doubling of dynamic level fluctuations. 2. Doubling of formant level differences. 3. Deepening of spectral minima (“zeros”). 4 Reduction of formant bandwidths. The most serious of these spectral degradations have been removed by a variable equalizer preceding the auto‐correlation analyzer. The remade speech quality produced by this technique exceeds that of known frequency‐channel vocoders and of a previously described cross‐correlation vocoder [J. Acoust. Soc. Am. 31, 852 (A) (1959)]. Bandwidth (or time) compression is about 10:1 (before multiplexing). Tape‐recorded speech samples from the equalized auto‐correlation vocoder will be played.

This content is only available via PDF.