Vector quantizer design for the coding of LSF parameters

Techniques for vector quantization of the line spectrum frequency (LSF) representation of speech spectra are described. Applications for spectral vector quantization include very low rate (300 to 1200 bits/s) speech coders, the half rate digital cellular speech coders, and many other linear prediction-based speech coders. Some of the techniques and tradeoffs for the quantization of LSF speech spectra for low data rate coding with minimal distortion are examined. The most important of these is the overlaying of binary trees on split-vector vector quantizing code books. The ultimate goal of this work is the implementation of speech coders on single-chip digital-signal-processing (DSP) architectures for real-time devices.<>