On using Kullback-Leibler distance to estimate vector quantization performance for line spectrum frequency parameters

In this paper we demonstrate that the Kullback-Leibler distance is a good approximation for the expected number of extra bits required to quantize the LSF vectors from a given set using a VQ based on a different set, without increasing the distortion. By using this result we give an analytical demonstration for an important property in VQ design that says that to improve the performance of a VQ, a larger training set is required. A method to determine the coding loss due to the split VQ is also presented.