Forward-backwards training of hybrid HMM/BN acoustic models.

Konstantin Markov, Satoshi Nakamura

Browse the full Interspeech paper archive.