Publication detail

Analysis of Multilingual BLSTM Acoustic Model on Low and High Resource Languages

KARAFIÁT, M. BASKAR, M. VESELÝ, K. GRÉZL, F. BURGET, L. ČERNOCKÝ, J.

Original Title

Analysis of Multilingual BLSTM Acoustic Model on Low and High Resource Languages

Type

conference paper

Language

English

Original Abstract

The paper provides an analysis of automatic speech recognition systems (ASR) based on multilingual BLSTM, where we used multi-task training with separate classification layer for each language. The focus is on low resource languages, where only a limited amount of transcribed speech is available. In such scenario, we found it essential to train the ASR systems in a multilingual fashion and we report superior results obtained with pre-trained multilingual BLSTM on this task. The high resource languages are also taken into account and we show the importance of language richness for multilingual training. Next, we present the performance of this technique as a function of amount of target language data. The importance of including context information into BLSTM multilingual systems is also stressed, and we report increased resilience of large NNs to overtraining in case of multi-task training.

Keywords

Automatic speech recognition, Multilingual neural networks, Bidirectional Long Short Term Memory

Authors

KARAFIÁT, M.; BASKAR, M.; VESELÝ, K.; GRÉZL, F.; BURGET, L.; ČERNOCKÝ, J.

Released

15. 4. 2018

Publisher

IEEE Signal Processing Society

Location

Calgary

ISBN

978-1-5386-4658-8

Book

Proceedings of ICASSP 2018

Pages from

5789

Pages to

5793

Pages count

5

URL

https://www.fit.vut.cz/research/publication/11720/

BibTex

@inproceedings{BUT155042,
  author="Martin {Karafiát} and Murali Karthick {Baskar} and Karel {Veselý} and František {Grézl} and Lukáš {Burget} and Jan {Černocký}",
  title="Analysis of Multilingual BLSTM Acoustic Model on Low and High Resource Languages",
  booktitle="Proceedings of ICASSP 2018",
  year="2018",
  pages="5789--5793",
  publisher="IEEE Signal Processing Society",
  address="Calgary",
  doi="10.1109/ICASSP.2018.8462083",
  isbn="978-1-5386-4658-8",
  url="https://www.fit.vut.cz/research/publication/11720/"
}

Documents

karafiat_icassp2018_0005789.pdf