Publication Details
Word-subword based keyword spotting with implications in OOV detection
Szőke Igor, Ing., Ph.D. (DCGM)
Hannemann Mirko, Ph.D.
Kombrink Stefan, Dipl.-Linguist.
speech recognition, keyword spotting, spoken term detection, OOV
The talk is on our work in designing hybrid word-subword keyword spotting systems, that maintain the accuracy of LVCSR, while allowing for detecting OOVs as sequences of sub-word units.
Main-stream systems for keyword spotting and spoken term detection are based on the series of Large Vocabulary Continuous Speech Recognizer with subsequent search in its output. These systems are limited by the vocabulary of the recognizer and are not able to detect Out of Vocabulary (OOV) words. This talk will present our work in designing hybrid word-subword keyword spotting systems, that maintain the accuracy of LVCSR, while allowing for detecting OOVs as sequences of sub-word units. We will also show the links of this work to the detection, description and clustering of OOVs, as investigated in the framework of the EC-sponsored project DIRAC.
@misc{BUT63577,
author="Jan {Černocký} and Igor {Szőke} and Mirko {Hannemann} and Stefan {Kombrink}",
title="Word-subword based keyword spotting with implications in OOV detection",
year="2010",
pages="34",
publisher="Institute of Electrical and Electronics Engineers",
address="Pacific Grove",
url="http://www.fit.vutbr.cz/research/groups/speech/publi/2010/asilomar_kwd_oov.ppt",
note="presentation, poster"
}