Speech recognition of spontaneous, noisy speech using auxiliary information in Bayesian networks