Comparative study of algorithms for VQ design using conventional and neural-net based approaches

Results are presented of a comparative study investigating the efficiency of the neural-net-based approaches (Kohonen and NNVQ) in comparison with the conventional (LBG and K-means) approaches for vector quantization. This study focuses on the accuracy and speed of these four methods for the VQ design problem using two different input sources: (1) Gauss Markov source, and (2) speech signal (digit strings). The results of the study show that both the LBG and NNVQ methods perform better than K-means and Kohonen in achieving more accurate vector quantization. Moreover, the NNVQ method offers computational advantages since the neural-net-based algorithm can be implemented with the use of parallel processors owing to its inherent parallelism