Publication result detail

Resources and Benchmarks for Keyword Search in Spoken Audio From Low-Resource Indian Languages

NADIMPALLI, V.; KESIRAJU, S.; BANKA, R.; KETHIREDDY, R.; GANGASHETTY, S.

Original Title

Resources and Benchmarks for Keyword Search in Spoken Audio From Low-Resource Indian Languages

English Title

Resources and Benchmarks for Keyword Search in Spoken Audio From Low-Resource Indian Languages

Type

WoS Article

Original Abstract

This paper presents the resources and benchmarks developed for keyword search (KWS)in spoken audio from six low-resource Indian languages (from two families), namely Gujarati, Hindi,Marathi, Odia, Tamil, and Telugu. The current work on constructing keywords and building benchmarkKWS systems is inspired by the popular IARPA Babel program and the subsequent works on low-resourceKWS. The keywords are constructed by taking into account their properties i.e., occurrence, length, andaverage confusability; and their effects on the evaluation metric - the term-weighted value (TWV).We makeuse of freely available speech datasets, and reprocess them to create resources for KWS, thereby addingvalue to the existing speech resources. Four ASR-based KWS systems are built, and their performance isanalyzed across the three keyword properties on all the six languages. The prepared keywords and otherrelated resources to replicate our experiments are made available for the public.We believe that the analysisand guidelines provided in this paper will not only help the research community, but also practitioners andengineers to easily create KWS resources for newer languages, datasets, and scenarios.

English abstract

This paper presents the resources and benchmarks developed for keyword search (KWS)in spoken audio from six low-resource Indian languages (from two families), namely Gujarati, Hindi,Marathi, Odia, Tamil, and Telugu. The current work on constructing keywords and building benchmarkKWS systems is inspired by the popular IARPA Babel program and the subsequent works on low-resourceKWS. The keywords are constructed by taking into account their properties i.e., occurrence, length, andaverage confusability; and their effects on the evaluation metric - the term-weighted value (TWV).We makeuse of freely available speech datasets, and reprocess them to create resources for KWS, thereby addingvalue to the existing speech resources. Four ASR-based KWS systems are built, and their performance isanalyzed across the three keyword properties on all the six languages. The prepared keywords and otherrelated resources to replicate our experiments are made available for the public.We believe that the analysisand guidelines provided in this paper will not only help the research community, but also practitioners andengineers to easily create KWS resources for newer languages, datasets, and scenarios.

Keywords

Keyword search, low-resource languages, term-weighted value (TWV)

Key words in English

Keyword search, low-resource languages, term-weighted value (TWV)

Authors

NADIMPALLI, V.; KESIRAJU, S.; BANKA, R.; KETHIREDDY, R.; GANGASHETTY, S.

RIV year

2023

Released

28.03.2022

ISBN

2169-3536

Periodical

IEEE Access

Volume

10

Number

2022

State

United States of America

Pages from

34789

Pages to

34799

Pages count

11

URL

https://ieeexplore.ieee.org/document/9743904

BibTex

@article{BUT182528,
  author="NADIMPALLI, V. and KESIRAJU, S. and BANKA, R. and KETHIREDDY, R. and GANGASHETTY, S.",
  title="Resources and Benchmarks for Keyword Search in Spoken Audio From Low-Resource Indian Languages",
  journal="IEEE Access",
  year="2022",
  volume="10",
  number="2022",
  pages="34789--34799",
  doi="10.1109/ACCESS.2022.3162854",
  issn="2169-3536",
  url="https://ieeexplore.ieee.org/document/9743904"
}

Documents

nadimpalli_IEEE-Access_2022_Resources_and_Benchmarks_for_Keyword_Search_in_Spoken_Audio_From_Low-Resource_Indian_Languages