Vector quantization using frequency-sensitive competitive-learning neural networks

A training algorithm is represented for a competitive learning network. This algorithm is applied to the problem of vector quantization using neural networks. An important advantage of using neural networks for vector quantization is that the computations can be carried out in parallel by the neural units. The performance of this algorithm is compared with other neural networks and traditional nonneural algorithms for vector quantization. The basic properties of the algorithm are discussed, the results of quantizing vectors of linear prediction coefficients from a speech signal are presented, and it is shown that the network yields results that are comparable to those obtained using the traditional algorithm.<>