Unsupervised speaker adaptation based on sufficient HMM statistics of selected speakers | IEEE Conference Publication | IEEE Xplore