Efficient coding of LPC parameters using adaptive prefiltering and MSVQ with partially adaptive codebook

The authors propose an efficient vector quantization scheme and a novel linear predictive coding (LPC) analysis scheme, both of which exploit interframe correlation in the successive spectrum envelope of speech signals. The first quantization scheme proposed is a multistage vector quantization of line spectrum pair (LSP) parameters with a partially adaptive codebook (MSVQ-AC). The second algorithm is an LPC analysis scheme, with closed-loop adaptive prefiltering (LPC-PF), which realized temporary higher-order analysis than the standard LPC with a few additional transmission bits. A combined system of the LPC-PF and two-split, two-stage VQ with the adaptive codebook can quantize tenth-order LSP parameters at around 23 bits/frame, realizing sufficient quality and reasonable complexity.<>