Dialect identification based on VQ codebook design with GA-LBG algorithm

Abstract

In order to solve the problem of GA's slow convergence in the VQ codebook design due to the strong global search ability and the complex computation, a hybrid algorithm based on the GA-LBG algorithm is adopted because the LBG algorithm as an iterative algorithm based on the nearest neighbor rule and the centroid rule owns the advantage of fast convergence. In the simulation experiment, MFCC extracted from Mandarin, Shanghainese, Cantonese and Hokkien are employed as the feature vectors to establish codebook models with GA-LBG for the dialect identification, and the recognition performances on different size of the VQ codebooks are studied. And simulation results demonstrate that the running time with GA-LBG reduces to 1066.4s, less than that with GA alone.