A new linear predictive method is presented in this study. The method, Linear Prediction with Linear Extrapolation (LPLE), reformulates the computation of linear prediction by combining the preceding values of sample x(n) into consecutive sample pairs (i.e., x(n-2i), x(n-2i+1)). Each of these pairs determines a regression line the value of which at time instant n is used as a data sample in the prediction. The optimal LPLE-predictor is obtained by minimizing the square of the prediction error using the autocorrelation method. The rationale for the new method is the fact that LPLE yields an all-pole filter of order 2p when the number of unknowns in the normal equations equals p. Therefore the new all-pole modeling method can be used in speech coding applications. Preliminary experiments of the present study show that LPLE is able to model speech spectra more accurately in comparison to conventional linear prediction in the case when a very small number of prediction parameters is required to be used in order to greatly compress the spectral information of speech signals.
Cite as: Alku, P., Varho, S. (1998) A new linear predictive method for compression of speech signals. Proc. 5th International Conference on Spoken Language Processing (ICSLP 1998), paper 0003, doi: 10.21437/ICSLP.1998-384
@inproceedings{alku98b_icslp,
author={Paavo Alku and Susanna Varho},
title={{A new linear predictive method for compression of speech signals}},
year=1998,
booktitle={Proc. 5th International Conference on Spoken Language Processing (ICSLP 1998)},
pages={paper 0003},
doi={10.21437/ICSLP.1998-384}
}