A New Adaptive Line Spectral Pair Filter for Linear Prediction of Speech.(Dept.E)

Abstract

Analysis/synthesis of speech by PARCOR method is proved to be advantageous over other methods previously developed for bit rates above 2.4 kbps, However, if the bit rate is reduced below this value, reconstructed speech becomes unclear and unnatural. Line spectral pair (LSP) coefficients were thus Introduced as alternative representations of linear prediction parameters for low-bit rate speech coding. This paper presents a least-mean-square (LMS) type adaptive LSP filter for directly calculating the LSP coefficients on sample-by-semple basis. The algorithm proposed here is shown to have higher convergence rate and lower misadjustment as compared to other algorithms.