7^{th} International Conference on Spoken Language ProcessingSeptember 16-20, 2002 |
The autocorrelation function of the all-pole filter given by the conventional linear prediction (LP) matches exactly the autocorrelation function of the input signal between indices 0 and m, when the prediction order equals m. This study describes a recently developed technique, Weighted-Sum Line Spectrum Pair (WLSP), where an allpole filter is defined by using a sum of weighted LSP (Line Spectrum Pair) polynomials. WLSP yields a stable all-pole filter of order m, whose autocorrelation function coincides to that of the input between indices 0 and m- 1. By sacrificing the exact matching of the autocorrelations at indexm, WLSP models the autocorrelation of the input at the indices above m more accurately than conventional LP. In the current paper, the performance of WLSP in spectral modelling of wide-band speech is analysed. It is shown that WLSP yields all-pole spectra that model the formant structure of vowels more accurately than conventional LP of the same prediction order.
Bibliographic reference. Alku, Paavo / Bäckström, Tom (2002): "All-pole modeling of wide-band speech using weighted sum of the LSP polynomials", In ICSLP-2002, 977-980.