Efficient LSP quantization algorithm for very low bit rate speech coding
Tang Kun · Journal of Tsinghua University(Science and Technology) · 2004
This paper presents an efficient linear spectrum pairs (LSP) parameters quantization algorithm called Pattern-based Residual Split Multi-Stage Matrix Quantization (P-RS-MSMQ). The method uses matrix quantization of multipleframes with not only mean-value-removal and inter-frame prediction basedon thesuper-frame pattern, but also matrix splitting and multi-stage matrix quantization. The algorithm also uses an intra-frame weighting scheme based on the energy of the short-time spectrum and an inter-frame weighting scheme based on the importance offrames in the super-frame. Simulation results demonstrate that the algorithm can achieve almost transparent LSP quantization at 700 b/s and good performance at the very low bit-rateof 300～400 b/s , so the algorithm is very useful for very low bit-rate speech coding.