Robust split vector quantization of lsp parameters at low bit rates

Abstract EN

In linear predictive coding algorithms, transmission of linear predictive coding (LPC) parameters, often transformed to the line spectrum pairs (LSP) representation, consumes a large part of the total bit rate of the coder.

At low bit rate, it is essential to quantize these parameters accurately using as few bits as possible without sacrificing the transparent quality and with low complexity.

Though the vector quantizers are more efficient than the scalar quantizers, their use for accurate quantization of LPC information is restricted due to their probable high complexity.

In this paper, a remedial solution to the problem of complexity is presented by comparing three structures of the split vector quantizer.

The 3–3– 4 split vector quantizer is found to be the best among the three schemes presented.

Two solutions to the problem of LSP inversion which may arise after split quantization are proposed.

In this paper three distance measures are tested with the split 3–3– 4 vector quantizer and the inverse harmonic mean (IHM) distance is found to be the most suitable with the best performance.

An algorithm is proposed for the effect of noise on the performance of the split 3–3– 4 quantizer and the results obtained are reported.