Improving semi-supervised deep neural network for keyword search in low resource languages

Roger Hsiao, Tim Ng, Le Zhang, Shivesh Ranjan, Stavros Tsakalidis, Long Nguyen, Richard Schwartz

In this work, we investigate how to improve semi-supervised DNN for low resource languages where the initial systems may have high error rate. We propose using semi-supervised MLP features for DNN training, and we also explore using confidence to improve semi-supervised cross entropy and sequence training. The work conducted in this paper was evaluated under the IARPA Babel program for the keyword spotting tasks. We focus on the limited condition where there are around 10 hours of supervised data for training.

doi: 10.21437/Interspeech.2014-280

Cite as: Hsiao, R., Ng, T., Zhang, L., Ranjan, S., Tsakalidis, S., Nguyen, L., Schwartz, R. (2014) Improving semi-supervised deep neural network for keyword search in low resource languages. Proc. Interspeech 2014, 1088-1091, doi: 10.21437/Interspeech.2014-280

@inproceedings{hsiao14_interspeech,
  author={Roger Hsiao and Tim Ng and Le Zhang and Shivesh Ranjan and Stavros Tsakalidis and Long Nguyen and Richard Schwartz},
  title={{Improving semi-supervised deep neural network for keyword search in low resource languages}},
  year=2014,
  booktitle={Proc. Interspeech 2014},
  pages={1088--1091},
  doi={10.21437/Interspeech.2014-280}
}