Detail výsledku
Query-by-example Spoken Term Detection Evaluation on Low-resource Languages
Rodriguez-Fuentes Luis, FIT (FIT)
Szőke Igor, Ing., Ph.D., UPGM (FIT)
Buzo Andi, FIT (FIT)
Metze Florian
a další
In this paper, besides presenting the setup, datasets and performancemeasures of the SWS 2013 evaluation, we have analyzedthe results obtained by the submitted systems and presented a postevaluationstudy where the 10 best-performing systems were incrementallyfused (at the score level), obtaining a 30%relative improvementover the best-performing individual system, proving the benefitsof combining independent or complementary sources of informationor different modeling approaches.Given the increasing interest for this task in the community, weare already planning a new edition of the SWS evaluation, renamedQUESST, i.e. Query by Example Spoken Search Task, within theMediaeval 2014 benchmark campaign. This year, we will continuetackling the problem of low-resource settings and will introduce acomponent of variability between queries and references, allowingfor a limited amount of acoustic insertions to still be consideredmatches.
benchmark evaluation, low-resource languages,query-by-example spoken term detection
@inproceedings{BUT111585,
author="Xavier {Anguera} and Luis {Rodriguez-Fuentes} and Igor {Szőke} and Andi {Buzo} and Florian {Metze}",
title="Query-by-example Spoken Term Detection Evaluation on Low-resource Languages",
booktitle="Proceedings of the 4th International Workshop on Spoken Language Technologies for Under- resourced Languages SLTU-2014. St. Petersburg, Russia",
year="2014",
pages="24--31",
publisher="International Speech Communication Association",
address="St. Petersburg",
isbn="978-5-8088-0908-6",
url="http://www.fit.vutbr.cz/research/groups/speech/publi/2014/anguera_sltu2014_p24_31.pdf"
}