Result Details
i-vectors in language modeling: An efficient way of domain adaptation for feed-forward models
Kesiraju Santosh, Ph.D., DCGM (FIT)
Burget Lukáš, doc. Ing., Ph.D., DCGM (FIT)
We show an effective way of adding context information toshallow neural language models. We propose to use SubspaceMultinomial Model (SMM) for context modeling and we addthe extracted i-vectors in a computationally efficient way. Byadding this information, we shrink the gap between shallowfeed-forward network and an LSTM from 65 to 31 points of perplexityon the Wikitext-2 corpus (in the case of neural 5-grammodel). Furthermore, we show that SMM i-vectors are suitablefor domain adaptation and a very small amount of adaptationdata (e.g. endmost 5% of a Wikipedia article) brings asubstantial improvement. Our proposed changes are compatiblewith most optimization techniques used for shallow feedforwardLMs.
language modeling, feed-forward models, subspacemultinomial model, domain adaptation
@inproceedings{BUT155102,
author="Karel {Beneš} and Santosh {Kesiraju} and Lukáš {Burget}",
title="i-vectors in language modeling: An efficient way of domain adaptation for feed-forward models",
booktitle="Proceedings of Interspeech 2018",
year="2018",
journal="Proceedings of Interspeech",
volume="2018",
number="9",
pages="3383--3387",
publisher="International Speech Communication Association",
address="Hyderabad",
doi="10.21437/Interspeech.2018-1070",
issn="1990-9772",
url="https://www.isca-speech.org/archive/Interspeech_2018/abstracts/1070.html"
}IT4Innovations excellence in science, MŠMT, Národní program udržitelnosti II, LQ1602, start: 2016-01-01, end: 2020-12-31, completed
Neural networks for signal processing and speech data mining, TAČR, Program na podporu aplikovaného výzkumu ZÉTA, TJ01000208, start: 2018-01-01, end: 2019-12-31, completed