Optimal scalar quantization of the LSP and the LAR for speech coding
Résumé
The Line Spectrum Pairs (LSP) provide an efficient representation of the synthesis filter used in Linear Predictive Coding (LPC) of speech. In this paper, several algorithms are studied for the quantization of these parameters including a new non uniform adaptive quantization forward method. The performance of the quantization methods of the LSP parameters is studied for a total number B of bits per frame varing from 10 to 40. The optimum bit allocation for each LSP is obtained by the Fox-Makhoul procedure. The proposed method for quantizing the LSP parameters is tested in a CELP coder running at 8 Kbits/s and compared to the optimum quantization of the LAR (Log Area Ratios) coefficients.