On the use of stochastic inference networks for representing multiple word pronunciations | IEEE Conference Publication | IEEE Xplore