An all-phoneme ergodic HMM for unsupervised speaker adaptation | IEEE Conference Publication | IEEE Xplore