Partial-energy weighted interpolation of linear prediction coefficients

This paper discusses the interpolation of linear prediction (LP) coefficients. The performance of LP analysis using different numbers of subframes and the choice of representation for the LP coefficients are studied. Interpolation is done by converting the LP coefficients in one of the following representations: line spectral frequencies, reflection coefficients, log area ratios, and autocorrelations. It is shown that good performance is obtained for line spectral frequencies and five subframes per frame. A new interpolation technique which incorporates partial frame energy is introduced. This technique generalizes the concept of energy weighting to different LP coefficient representations.