Line Spectral Frequencies (LSF) are widely used to quantize spectral information in most of low-bit-rate linear predictive speech coders because of their desirable features with regard to ordering property and spectral sensitivity. Although LSF vector quantizer uses fewer bits , scalar quantization of LSF  is still adopted because codebook storage and computational complexity are the major considerations for real-time implementation. In this paper, a new set of LSF-based coefficients is proposed which can be coded efficiently by increasing the utilization of quantization table. With the novel table search technique based on vector measure, 10th-order LPC parameters can be coded at 28 bits/frame with transparent quantization quality.
Bibliographic reference. Ng, H. C. / Leung, S. H. (1995): "Scalar quantization of LSF parameters at 28 bits/frame", In EUROSPEECH-1995, 1061-1064.