Publication Details
Score Normalization and System Combination for Improved Keyword Spotting
Schwartz Richard (FIT)
Tsakalidis Stavros (FIT)
Zhang Le (FIT)
Ranjan Shivesh (FIT)
Ng Tim
Hsiao Roger (FIT)
Nguyen Long
Grézl František, Ing., Ph.D. (DCGM)
Hannemann Mirko, Ph.D.
Karafiát Martin, Ing., Ph.D. (DCGM)
Szőke Igor, Ing., Ph.D. (DCGM)
Veselý Karel, Ing., Ph.D. (DCGM)
and others
keyword search, score normalization, system combination, indexing and search
This article is about Score Normalization and System Combination for Improved Keyword Spotting (KWS).
We present two techniques that are shown to yield improved Key- word Spotting (KWS) performance when using the ATWV/MTWV performance measures: (i) score normalization, where the scores of different keywords become commensurate with each other and they more closely correspond to the probability of being correct than raw posteriors; and (ii) system combination, where the detections of multiple systems are merged together, and their scores are in- terpolated with weights which are optimized using MTWV as the maximization criterion. Both score normalization and system com- bination approaches show that significant gains in ATWV/MTWV can be obtained, sometimes on the order of 8-10 points (absolute), in five different languages. A variant of these methods resulted in the highest performance for the official surprise language evaluation for the IARPA-funded Babel project in April 2013.
@inproceedings{BUT105974,
author="Damianos {Karakos} and Richard {Schwartz} and Stavros {Tsakalidis} and Le {Zhang} and Shivesh {Ranjan} and Tim {Ng} and Roger {Hsiao} and Long {Nguyen} and František {Grézl} and Mirko {Hannemann} and Martin {Karafiát} and Igor {Szőke} and Karel {Veselý}",
title="Score Normalization and System Combination for Improved Keyword Spotting",
booktitle="Proceedings of ASRU 2013",
year="2013",
pages="210--215",
publisher="IEEE Signal Processing Society",
address="Olomouc",
doi="10.1109/ASRU.2013.6707731",
isbn="978-1-4799-2755-5",
url="https://ieeexplore.ieee.org/document/6707731"
}