Publication Details

Multi-task Neural Networks For Speech Recognition

EGOROVA Ekaterina. Multi-task Neural Networks For Speech Recognition. In: Proceedings of the 20th Student Conference, EEICT 2014. Volume 2. Brno: Brno University of Technology, 2014, pp. 24-26. ISBN 978-80-214-4923-7. Available from: http://www.feec.vutbr.cz/EEICT/wp-content/uploads/2014/04/2014-sbornik-mgr.pdf
Czech title
Víceúkolové trénování neuronových sítí pro rozpoznávání řeči
Type
conference paper
Language
english
Authors
Egorova Ekaterina, Ing., Ph.D. (DCGM FIT BUT)
URL
Keywords

Speech recognition, neural networks, deep neural networks, multi-task neural networks.

Annotation

The article covers experiments on TIMIT database exploring the possibility of using multitask neural networks for speech recognition. Multi-task neural networks are deep neural networks solving several different classification tasks simultaneously. The secondary tasks chosen for the experiments are gender, context, articulatory characteristics and a fusion of some of them. The experiments show that addition of such tasks can enhance the learning and improve recognition accuracy.

Published
2014
Pages
24-26
Proceedings
Proceedings of the 20th Student Conference, EEICT 2014
Series
Volume 2
Conference
Student EEICT 2014, Brno, CZ
ISBN
978-80-214-4923-7
Publisher
Brno University of Technology
Place
Brno, CZ
BibTeX 
@INPROCEEDINGS{FITPUB10629,
   author = "Ekaterina Egorova",
   title = "Multi-task Neural Networks For Speech Recognition",
   pages = "24--26",
   booktitle = "Proceedings of the 20th Student Conference, EEICT 2014",
   series = "Volume 2",
   year = 2014,
   location = "Brno, CZ",
   publisher = "Brno University of Technology",
   ISBN = "978-80-214-4923-7",
   language = "english",
   url = "https://www.fit.vut.cz/research/publication/10629"
}
Back to top