Title :
Correcting spectral envelope shifts in linear predictive speech compression systems
Author :
Nguyen, Vien ; Goncharoff, V. ; Damoulakis, John
Author_Institution :
Martin Marietta Aero & Naval Syst., Glen Burnie, MD, USA
fDate :
30 Sep-3 Oct 1990
Abstract :
An algorithm to correct for spectral envelope distortion in the synthetic excitation signal in residual excited linear-predictive (LP)-based vocoders is presented. In these coders, if the residual is regenerated exactly, the original signal can be recovered exactly. However, exact recovery of the residual is not possible since it is always quantized by some sort of coding process, causing it always to have a different spectral envelope than the original. The result is an output signal spectrum not matching that of the input signal. To correct for this kind of spectral distortion, the proposed algorithm modifies the LPC spectral envelope by the amount the coded residual differs from being white. The modification requires only a minimum additional amount of computations that can be greatly simplified if the spectral estimation is computed using autocorrelation techniques
Keywords :
encoding; filtering and prediction theory; spectral analysis; speech analysis and processing; speech synthesis; vocoders; LPC spectral envelope; autocorrelation techniques; coded residual; coding; input signal; linear predictive vocoders; output signal spectrum; residual excited linear prediction; spectral envelope distortion; spectral envelope shifts correction; spectral estimation; speech compression systems; synthetic excitation signal; Autocorrelation; Decoding; Equations; Filtering; Filters; Linear predictive coding; Quantization; Speech coding; Speech synthesis; Vocoders;
Conference_Titel :
Military Communications Conference, 1990. MILCOM '90, Conference Record, A New Era. 1990 IEEE
Conference_Location :
Monterey, CA
DOI :
10.1109/MILCOM.1990.117441
