VQ codebook design using hybrid ant colony algorithm for speech line spectral frequencies

A significant issue in vector quantization (VQ) is the design of the codebook. The standard method for codebook design has been the GLA algorithm and Lind, Buzo and Gray (LBG) algorithm. Theses algorithms can get stuck in suboptimal codebooks due to the presence of several locally minimum distortion values. Simulated annealing (SA) is an optimization procedure that uses randomness to escape local minima in its search for a globally minimum state. Ant colony algorithm is a stochastic searching optimization algorithm. In this paper, an appropriately adapted ant colony system (ACS) combined with simulated annealing (SA) and Enhanced GLA (EGLA) are proposed for VQ codebook design. The results presented for speech samples represented by line spectral frequencies (LSF) indicate that the resulting design with hybrid ant colony with simulated annealing give the best results compared to GLA and hybrid ant colony with EGLA algorithms in terms of distortion and execution time.