Data Augmentation Using Spectral Warping for Low Resource Children ASR

dc.contributorAalto-yliopistofi
dc.contributorAalto Universityen
dc.contributor.authorKathania, Hemant Kumar
dc.contributor.authorKadyan, Viredner
dc.contributor.authorKadiri, Sudarsana Reddy
dc.contributor.authorKurimo, Mikko
dc.contributor.departmentSpeech Recognition
dc.contributor.departmentUniversity of Petroleum and Energy Studies
dc.contributor.departmentSpeech Communication Technology
dc.contributor.departmentDept Signal Process and Acoust
dc.date.accessioned2022-11-23T08:04:18Z
dc.date.available2022-11-23T08:04:18Z
dc.date.issued2022-12
dc.descriptionFunding Information: This work was supported by the Academy of Finland (grants 329267, 330139). Publisher Copyright: © 2022, The Author(s).
dc.description.abstractIn low resource children automatic speech recognition (ASR) the performance is degraded due to limited acoustic and speaker variability available in small datasets. In this paper, we propose a spectral warping based data augmentation method to capture more acoustic and speaker variability. This is carried out by warping the linear prediction (LP) spectra computed from speech data. The warped LP spectra computed in a frame-based manner are used with the corresponding LP residuals to synthesize speech to capture more variability. The proposed augmentation method is shown to improve the ASR system performance over the baseline system. We have compared the proposed method with four well-known data augmentation methods: pitch scaling, speaking rate, SpecAug and vocal tract length perturbation (VTLP), and found that the proposed method performs the best. Further, we have combined the proposed method with these existing data augmentation methods to improve the ASR system performance even more. The combined system consisting of the original data, VTLP, SpecAug and the proposed spectral warping method gave the best performance by a relative word error rate reduction of 32.13% and 10.51% over the baseline system for Punjabi children and TLT-school corpus, respectively. The proposed spectral warping method is publicly available at https://github.com/kathania/Spectral-Warping.en
dc.description.versionPeer revieweden
dc.format.extent7
dc.format.mimetypeapplication/pdf
dc.identifier.citationKathania , H K , Kadyan , V , Kadiri , S R & Kurimo , M 2022 , ' Data Augmentation Using Spectral Warping for Low Resource Children ASR ' , Journal of Signal Processing Systems , vol. 94 , no. 12 , pp. 1507-1513 . https://doi.org/10.1007/s11265-022-01820-0en
dc.identifier.doi10.1007/s11265-022-01820-0
dc.identifier.issn1939-8018
dc.identifier.otherPURE UUID: ffc05710-eff6-4a51-b442-792e41683c23
dc.identifier.otherPURE ITEMURL: https://research.aalto.fi/en/publications/ffc05710-eff6-4a51-b442-792e41683c23
dc.identifier.otherPURE LINK: http://www.scopus.com/inward/record.url?scp=85141370357&partnerID=8YFLogxK
dc.identifier.otherPURE LINK: https://link.springer.com/content/pdf/10.1007/s11265-022-01820-0.pdf
dc.identifier.otherPURE FILEURL: https://research.aalto.fi/files/92753758/Kathania_et_alii_Data_Augmentation_Using_Spectral_Warping_for_Low_Resource_Children_ASR.pdf
dc.identifier.urihttps://aaltodoc.aalto.fi/handle/123456789/117893
dc.identifier.urnURN:NBN:fi:aalto-202211236653
dc.language.isoenen
dc.publisherSPRINGER
dc.relation.ispartofseriesJournal of Signal Processing Systemsen
dc.rightsopenAccessen
dc.subject.keywordChildren speech recognition
dc.subject.keywordProsody modification
dc.subject.keywordSpecAug
dc.subject.keywordSpectral warping
dc.subject.keywordTDNN
dc.subject.keywordVTLP
dc.titleData Augmentation Using Spectral Warping for Low Resource Children ASRen
dc.typeA1 Alkuperäisartikkeli tieteellisessä aikakauslehdessäfi
dc.type.versionpublishedVersion
Files