Sequence-discriminative training of deep neural networks

Sequence-discriminative training of deep neural networks (DNNs) is investigated on a standard 300 hour American English conversational telephone speech task.Different sequencediscriminative criteria -maximum mutual information (MMI), minimum phone error (MPE), state-level minimum Bayes risk (sMBR), and boosted MMI -are compared.Two different heuristics are investigated to improve the performance of the DNNs trained using sequence-based criteria -lattices are regenerated after the first iteration of training; and, for MMI and BMMI, the frames where the numerator and denominator hypotheses are disjoint are removed from the gradient computation.Starting from a competitive DNN baseline trained using cross-entropy, different sequence-discriminative criteria are shown to lower word error rates by 7-9% relative, on average.Little difference is noticed between the different sequencebased criteria that are investigated.The experiments are done using the open-source Kaldi toolkit, which makes it possible for the wider community to reproduce these results.