ON THE USE OF GRAPHEME MODELS FOR SEARCHING IN LARGE SPOKEN
ARCHIVES

D 2018

ON THE USE OF GRAPHEME MODELS FOR SEARCHING IN LARGE SPOKEN ARCHIVES

SVEC, Jan, Josef V. PSUTKA, Jan TRMAL, Lubos SMIDL, Pavel IRCING et. al.

Basic information

Original name

ON THE USE OF GRAPHEME MODELS FOR SEARCHING IN LARGE SPOKEN ARCHIVES

Authors

SVEC, Jan (203 Czech Republic), Josef V. PSUTKA (203 Czech Republic), Jan TRMAL (203 Czech Republic), Lubos SMIDL (203 Czech Republic), Pavel IRCING (203 Czech Republic) and Jan SEDMIDUBSKÝ (203 Czech Republic, guarantor, belonging to the institution)

Edition

Neuveden, 43rd IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP 2018), p. 6259-6263, 5 pp. 2018

Publisher

IEEE Computer Society

Other information

Language

English

Type of outcome

Stať ve sborníku

Field of Study

10201 Computer sciences, information science, bioinformatics

Confidentiality degree

není předmětem státního či obchodního tajemství

Publication form

storage medium (CD, DVD, flash disk)

RIV identification code

RIV/00216224:14330/18:00100802

Organization unit

Faculty of Informatics

ISBN

978-1-5386-4658-8

ISSN

DOI

http://dx.doi.org/10.1109/ICASSP.2018.8461774

UT WoS

000446384606084

Keywords in English

spoken term detection; speech indexing; grapheme-based speech recognition; keyword search

Abstract

V originále

This paper explores the possibility to use grapheme-based word and sub-word models in the task of spoken term detection (STD). The usage of grapheme models eliminates the need for expert-prepared pronunciation lexicons (which are often far from complete) and/or trainable grapheme-to-phoneme (G2P) algorithms that are frequently rather inaccurate, especially for rare words (words coming from a different language). Moreover, the G2P conversion of the search terms that need to be performed on-line can substantially increase the response time of the STD system. Our results show that using various grapheme-based models, we can achieve STD performance (measured in terms of ATWV) comparable with phoneme-based models but without the additional burden of G2P conversion.

Links

GBP103/12/G084, research and development project

Name: Centrum pro multi-modální interpretaci dat velkého rozsahu

Investor: Czech Science Foundation

Citovat

SVEC, Jan, Josef V. PSUTKA, Jan TRMAL, Lubos SMIDL, Pavel IRCING and Jan SEDMIDUBSKÝ. ON THE USE OF GRAPHEME MODELS FOR SEARCHING IN LARGE SPOKEN ARCHIVES. In 43rd IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP 2018). Neuveden: IEEE Computer Society, 2018, p. 6259-6263. ISBN 978-1-5386-4658-8. Available from: https://dx.doi.org/10.1109/ICASSP.2018.8461774.

@inproceedings{1405237,
   author = {Svec, Jan and Psutka, Josef V. and Trmal, Jan and Smidl, Lubos and Ircing, Pavel and Sedmidubský, Jan},
   address = {Neuveden},
   booktitle = {43rd IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP 2018)},
   doi = {http://dx.doi.org/10.1109/ICASSP.2018.8461774},
   keywords = {spoken term detection; speech indexing; grapheme-based speech recognition; keyword search},
   howpublished = {paměťový nosič},
   language = {eng},
   location = {Neuveden},
   isbn = {978-1-5386-4658-8},
   pages = {6259-6263},
   publisher = {IEEE Computer Society},
   title = {ON THE USE OF GRAPHEME MODELS FOR SEARCHING IN LARGE SPOKEN ARCHIVES},
   year = {2018}
}

TY  - JOUR
ID  - 1405237
AU  - Svec, Jan - Psutka, Josef V. - Trmal, Jan - Smidl, Lubos - Ircing, Pavel - Sedmidubský, Jan
PY  - 2018
TI  - ON THE USE OF GRAPHEME MODELS FOR SEARCHING IN LARGE SPOKEN ARCHIVES
PB  - IEEE Computer Society
CY  - Neuveden
SN  - 9781538646588
KW  - spoken term detection
KW  - speech indexing
KW  - grapheme-based speech recognition
KW  - keyword search
N2  - This paper explores the possibility to use grapheme-based word and sub-word models in the task of spoken term detection (STD). The usage of grapheme models eliminates the need for expert-prepared pronunciation lexicons (which are often far from complete) and/or trainable grapheme-to-phoneme (G2P) algorithms that are frequently rather inaccurate, especially for rare words (words coming from a different language). Moreover, the G2P conversion of the search terms that need to be performed on-line can substantially increase the response time of the STD system. Our results show that using various grapheme-based models, we can achieve STD performance (measured in terms of ATWV) comparable with phoneme-based models but without the additional burden of G2P conversion.
ER  -

SVEC, Jan, Josef V. PSUTKA, Jan TRMAL, Lubos SMIDL, Pavel IRCING and Jan SEDMIDUBSKÝ. ON THE USE OF GRAPHEME MODELS FOR SEARCHING IN LARGE SPOKEN ARCHIVES. In \textit{43rd IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP 2018)}. Neuveden: IEEE Computer Society, 2018, p.~6259-6263. ISBN~978-1-5386-4658-8. Available from: https://dx.doi.org/10.1109/ICASSP.2018.8461774.

Detailed Information on Publication Record