2016
Data selection from multiple ASR systems' hypotheses for unsupervised acoustic model training
ICASSP 2016accepted
This paper addresses unsupervised training of DNN acoustic model, by exploiting a large amount of unlabeled data with CRF-based classifiers. In the proposed scheme, we obtain ASR hypotheses by complementary GMM and DNN based ASR systems. Then, a set of dedicated classifiers are designed and trained…