Result Details
Adaptation of Multilingual Stacked Bottle-neck Neural Network Structure for New Language
Karafiát Martin, Ing., Ph.D., DCGM (FIT)
Veselý Karel, Ing., Ph.D., DCGM (FIT)
In this paper a multilingual training of Stacked Bottle-Neck neural network structure for feature extraction is addressed. While for languageswith plentiful resources, the optimal approach is to train theBN-NN on the target data, limited resources call for re-using datafrom other languages.
feature extraction, Bottle-neck features, neuralnetwork adaptation, multilingual neural networks, Stacked Bottle-Neck structure
The neural network based features became an inseparable part of state-of-the-art LVCSR systems. In order to perform well, the network has to be trained on a large amount of in-domain data. With the increasing emphasis on fast development of ASR system on limited resources, there is an effort to alleviate the need of in-domain data. To evaluate the effectiveness of other resources, we have trained the Stacked Bottle-Neck neural networks structure on multilingual data investigating several training strategies while treating the target language as the unseen one. Further, the systems were adapted to the target language by re-training. Finally, we evaluated the effect of adaptation of individual NNs in the Stacked Bottle-Neck structure to find out the optimal adaptation strategy. We have shown that the adaptation can significantly improve system performance over both, the multilingual network and network trained only on target data. The experiments were performed on Babel Year 1 data.
@inproceedings{BUT111544,
author="František {Grézl} and Martin {Karafiát} and Karel {Veselý}",
title="Adaptation of Multilingual Stacked Bottle-neck Neural Network Structure for New Language",
booktitle="Proceedings of ICASSP 2014",
year="2014",
pages="7704--7708",
publisher="IEEE Signal Processing Society",
address="Florencie",
doi="10.1109/ICASSP.2014.6855089",
isbn="978-1-4799-2892-7",
url="http://www.fit.vutbr.cz/research/groups/speech/publi/2014/grezl_icassp2014_p7704_adapation.pdf"
}IARPA Building Speech Recognition for Keyword Search in a New Language in a Week with Limited Training Data (BABEL) - Babelon, BBN, start: 2012-03-05, end: 2016-11-04, completed
Speech recognition for low-resource languages, GACR, Postdoktorandské granty, GPP202/12/P604, start: 2012-01-01, end: 2014-12-31, completed
Zpracování, rozpoznávání a zobrazování multimediálních a 3D dat, BUT, Vnitřní projekty VUT, FIT-S-14-2506, start: 2014-01-01, end: 2016-12-31, completed