GMM based speaker identification using training-time-dependent number of mixtures | IEEE Conference Publication | IEEE Xplore