Efficient interblock noiseless coding of speech LPC parameters

Abstract

Many speech coding systems apply the principles of vector quantization (VQ) to the spectral information since VQ is much more efficient than simple scalar quantization. Apart from ordinary VQ for very-low-rate speech coding, split VQ can be used in high-quality coding systems for transparent coding of speech LPC parameters. This paper proposes relative index coding (RIC) as a new principle for efficient entropy coding of LPC parameters. RIC can be combined with various VQ procedures increasing the coding efficiency by exploiting correlations between successive VQ indices. In combination with conventional VQ, RIC yields bit savings of up to 30%. Combined with split VQ, it reduces the average bit rate necessary for transparent coding from 24 bits/frame down to 175 bits/frame at a frame period of 10 ms. Further savings of more than 0.5 bits/frame can be obtained by introducing an entropy-constrained split VQ. >