Improvement of Delayed Decision Coding for LSF Difference Quantization

The line spectrum frequencies (LSFs) are significant parameters which characterize the spectrum of the speech. The differential quantization of LSFs has been proven to provide better performance which exploits the intraframe correlation between adjacent LSFs, and the delayed decision coding of LSF differences (LSFDs) has been developed to obtain globally optimal quantization. In this paper, we improve the delayed decision quantization of LSFDs by using the conditional codebooks. Test results show that the algorithm proposed can perform transparent quantization of LSFs with 28 bits/frame.