Kodierungsverfahren und Sprachkodierer unter Anwendung von Analyse durch lineare Prädiktion

摘要

Method of speech coding which uses a model of production of the word by passing an excitation signal Ck, representing the vocal source and subjected to amplification G, through a long term predictor filter of transfer function 1/B(z) where B(z)=1-bz<-T>, and through a short term predictor filter of transfer function 1/A(z)= SIGMA aiz representing the contribution of the vocal duct. Each frame is represented by values of the parameters Ck, G, ai, T and b. An error signal is generated by subtracting, from the frame of the original word signal, two terms. The said word signal frame is, after subtraction, subjected to short-term analysis filtering and to a perceptual weighted synthesis filtering H, the first term being representative of the output of the predictor in the long term with the time offset T, subjected to the synthesis filtering, while the second term is representative of each of the excitation signals in its turn. Each signal is subjected in advance to amplification G and to the same weighted synthesis filtering as the frame of the word signal. In the course of the same sequence an optimum of the shift T is determined by looking for a perceptual error minimum and the coefficient k is deduced from it, then the optimal values of b and of G are calculated.