Fast search algorithms with applications to split and multi-stage vector quantization of speech lsp parameters
Sunil Arya, Nam Phamdo, Nariman Farvardin, David M. Mount · 2005
Linear predictive coding (LPC) is an indispensable component of most speech coding algorithms. Thus, efficient representation of the LPC parameters is a central issue in speech coding. There is evidence that the most efficient representation of the LPC parameters is the line spectrum pair (LSP) representation [l]. The LSP parameters have two properties which make them amenable to quantization [l]: (1) their ordering property and (2) their interpolation property. It is generally believed that, due to these properties, for the same quantization scheme, quantizers which operate on LSP parameters are more efficient than quantizers which operate on other LPC representations. Results of several studies [l, 21 support this belief.