This paper describes the keyword search system developed by the STC
team in the framework of OpenKWS 2016 evaluation. The acoustic modeling
techniques included i-vectors based speaker adaptation, multilingual
speaker-dependent bottleneck features, and a combination of feedforward
and recurrent neural networks. To improve the language model, we augmented
the training data provided by the organizers with texts generated by
the character-level recurrent neural networks trained on different
data sets. This led to substantial reductions in the out-of-vocabulary
(OOV) and word error rates. The OOV search problem was solved with
the help of a novel approach based on lattice generated phone posteriors
and a highly optimized decoder. This approach outperformed familiar
OOV search implementations in terms of speed and demonstrated comparable
or better search quality.
The system was among
the top three systems in the evaluation.
Cite as: Khokhlov, Y., Medennikov, I., Romanenko, A., Mendelev, V., Korenevsky, M., Prudnikov, A., Tomashenko, N., Zatvornitsky, A. (2017) The STC Keyword Search System for OpenKWS 2016 Evaluation. Proc. Interspeech 2017, 3602-3606, doi: 10.21437/Interspeech.2017-1212
@inproceedings{khokhlov17b_interspeech,
author={Yuri Khokhlov and Ivan Medennikov and Aleksei Romanenko and Valentin Mendelev and Maxim Korenevsky and Alexey Prudnikov and Natalia Tomashenko and Alexander Zatvornitsky},
title={{The STC Keyword Search System for OpenKWS 2016 Evaluation}},
year=2017,
booktitle={Proc. Interspeech 2017},
pages={3602--3606},
doi={10.21437/Interspeech.2017-1212}
}