Deep neural network (DNN) acoustic models yield posterior probabilities of senone classes. Recent studies support the existence of low-dimensional subspaces underlying senone posteriors. Principal component analysis (PCA) is applied to identify eigenposteriors and perform low-dimensional projection of the training data posteriors. The resulted enhanced posteriors are applied as soft targets for training better DNN acoustic model under the student-teacher framework. The present work advances this approach by studying incorporation of sequence discriminative training. We demonstrate how to combine the gains from eigenposterior based enhancement with sequence discrimination to improve ASR using semi-supervised training. Evaluation on AMI meeting corpus yields nearly 4% absolute reduction in word error rate (WER) compared to the baseline DNN trained with cross entropy objective. In this context, eigenposterior enhancement of the soft targets is crucial to enable additive improvement using out-of-domain untranscribed data.
Cite as: Dighe, P., Asaei, A., Bourlard, H. (2017) Exploiting Eigenposteriors for Semi-Supervised Training of DNN Acoustic Models with Sequence Discrimination. Proc. Interspeech 2017, 3552-3556, doi: 10.21437/Interspeech.2017-1784
@inproceedings{dighe17_interspeech,
author={Pranay Dighe and Afsaneh Asaei and Hervé Bourlard},
title={{Exploiting Eigenposteriors for Semi-Supervised Training of DNN Acoustic Models with Sequence Discrimination}},
year=2017,
booktitle={Proc. Interspeech 2017},
pages={3552--3556},
doi={10.21437/Interspeech.2017-1784}
}