Efficient LSP computation and quantization

Linear predictive coding (LPC) coefficients are widely used in a various speech coding method. Line spectrum pairs (LSP), as alternate representation of LPC presents some useful properties for quantization and transmission. This work investigates the computing algorithm of LSP and proposes an efficient method to isolate the LSP in pairs of roots in order to reduce the required computational complexity for root finding. A combined LSP computation and quantization are proposed and various correction techniques are investigated, in order to preserve the LSP order properties after quantization.