TRAP based features for LVCSR of meeting data

Czech title

TRAP příznaky pro rozponávání řeči aplikované na meeting data

Type

conference paper

Language

English

Authors

Karafiát Martin, Ing., Ph.D. (DCGM)
Grézl František, Ing., Ph.D. (DCGM)
Černocký Jan, prof. Dr. Ing. (DCGM)

URL

http://www.fit.vutbr.cz/~karafiat/publi/2004/karafiat_icslp2004.pdf

Keywords

Speech recognition, TRAP, LVCSR

Abstract

This paper describes using temporal patterns (TRAPs) feature extraction in large vocabulary continuous speech recognition (LVCSR) of meeting data. Frequency differentiation and local operators are applied to critical-band speech spectrum. Tests are performed with HMM recognizer on ICSI meetings database. We show that TRAP features in combination with standard ones lead to improvement of word-error rate (WER).

Annotation

This paper describes using temporal patterns (TRAPs) feature extraction in large vocabulary continuous speech recognition (LVCSR) of meeting data. Frequency differentiation and local operators are applied to critical-band speech spectrum. Tests are performed with HMM recognizer on ICSI meetings database. We show that TRAP features in combination with standard ones lead to improvement of word-error rate (WER).

Published

2004

Pages

437–440

Journal

8th International Conference on Spoken Language Processing, vol. 2004, no. 10, ISSN 1225-4111

Proceedings

Proc. 8th International Conference on Spoken Language Processing

Publisher

Sunjin Printing Co,

Place

Jeju Island

BibTeX

@inproceedings{BUT17131,
  author="Martin {Karafiát} and František {Grézl} and Jan {Černocký}",
  title="TRAP based features for LVCSR of meeting data",
  booktitle="Proc. 8th International Conference on Spoken Language Processing",
  year="2004",
  journal="8th International Conference on Spoken Language Processing",
  volume="2004",
  number="10",
  pages="437--440",
  publisher="Sunjin Printing Co,",
  address="Jeju Island",
  issn="1225-4111",
  url="http://www.fit.vutbr.cz/~karafiat/publi/2004/karafiat_icslp2004.pdf"
}