An Algorithm for Fast Direct Calculation of Quantized Lsp Parameters

Line Spectrum Pairs (LSP) are commonly used in speech coding for spectral quantization. Usually, the LSPs are first calculated, and then quantized. When scalar quantization is used, computational saving can be achieved by direct calculation of the quantized LSPs, searching zero-crossings on the grid formed by the values of the quantization tables (quantized-domain search). In a previous publication we have shown that by combining Kabal’s LSP calculation method with a quantized-domain search on a 34-bit quantization table, the number of polynomial evaluations is reduced from 150 to 71. In this paper we consider the use of a binary-tree quantizeddomain search to further decrease the complexity. It is explained why this type of search cannot be combined with Kabal’s method, and it is shown that Saoudi's LSP calculation method can be easily combined with a binary-tree quantized-domain search, obtaining a fast algorithm for direct quantized LSP calculation. The proposed algorithm reduces computational complexity and improves quantization performance.