Fast Speaker Adaption Via Maximum Penalized Likelihood Kernel Regression | IEEE Conference Publication | IEEE Xplore