Fourth European Conference on Speech Communication and Technology

Madrid, Spain
September 18-21, 1995

LSP Markov Model for Reducing the Complexity of Vector Quantization

B. Kovesi (1,2), S. Saoudi (1), J. M. Boucher (1), Z. Reguly (2)

(1) ENSTBr, Dept. SC, Technopole de Brest Iroise, Brest, France
(2) Technical University of Budapest, Dept. MMT., Budapest, Hungary

In this paper the vector quantization (VQ) of the LSP (Line Spectrum Pairs) coefficients of the CELP (Code Excited Linear Predictive) coder is addressed. Methods to reduce the complexity of VQ are proposed. First a method, which determines the search zone for the input vector's nearest neighbour in the codebook by comparing the norm of the vectors, is presented. This is followed by the description of a method based on the correlation of consecutive quantized LSP vectors. Here the transition matrix of the n states Markov model is used to reduce the searching zone for the nearest neighbour. The experimental results show that the combination of these two methods is an effective way of reducing the complexity of the vector quantization.

Full Paper

Bibliographic reference.  Kovesi, B. / Saoudi, S. / Boucher, J. M. / Reguly, Z. (1995): "LSP Markov model for reducing the complexity of vector quantization", In EUROSPEECH-1995, 1041-1044.