Publication result 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

English Title

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

Type

Paper in proceedings (conference paper)

Original Abstract

The paper provides an analysis of automatic speech recognitionsystems (ASR) based on multilingual BLSTM, where weused multi-task training with separate classification layer foreach language. The focus is on low resource languages, whereonly a limited amount of transcribed speech is available. Insuch scenario, we found it essential to train the ASR systemsin a multilingual fashion and we report superior resultsobtained with pre-trained multilingual BLSTM on this task.The high resource languages are also taken into account andwe show the importance of language richness for multilingualtraining. Next, we present the performance of this techniqueas a function of amount of target language data. The importanceof including context information into BLSTM multilingualsystems is also stressed, and we report increased resilienceof large NNs to overtraining in case of multi-tasktraining.

English abstract

The paper provides an analysis of automatic speech recognitionsystems (ASR) based on multilingual BLSTM, where weused multi-task training with separate classification layer foreach language. The focus is on low resource languages, whereonly a limited amount of transcribed speech is available. Insuch scenario, we found it essential to train the ASR systemsin a multilingual fashion and we report superior resultsobtained with pre-trained multilingual BLSTM on this task.The high resource languages are also taken into account andwe show the importance of language richness for multilingualtraining. Next, we present the performance of this techniqueas a function of amount of target language data. The importanceof including context information into BLSTM multilingualsystems is also stressed, and we report increased resilienceof large NNs to overtraining in case of multi-tasktraining.

Keywords

Automatic speech recognition, Multilingualneural networks, Bidirectional Long Short Term Memory

Key words in English

Automatic speech recognition, Multilingualneural networks, Bidirectional Long Short Term Memory

Authors

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

RIV year

2019

Released

15.04.2018

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