Vector quantization of LSP parameters by split

Abstract

An important problem in speech coding is the quantization of the linear predictive coefficients (LPC) with the smallest possible number of bits. Since the direct quantization of LPC coefficients is known to be unsatisfactory, we consider this problem for an equivalent representation namely the line spectrum pairs (LSPs). The performance of two algorithms for the quantization of the LSP parameters is studied. For the first algorithm, the well-known ordering property of LSP parameters is exploited in quantizing the frequency differences of consecutive LSP parameters. Vector quantizers are more efficient than scalar quantizers, their use for accurate quantization of LSP parameters is impeded due to their prohibitively high complexity problem. A split vector quantizer is used to overcome the complexity problem. The LPC vector consisting of 10 LSPs is split into three parts. For the spectral distortion measure, appropriate performance comparisons between the two quantization schemes are rendered.