Detailed Information

Cited 0 time in webofscience Cited 0 time in scopus
Metadata Downloads

Sequence dicriminative training 기법을 사용한 트랜스포머 기반 음향 모델 성능 향상

Full metadata record
DC Field Value Language
dc.contributor.author이채원-
dc.contributor.author장준혁-
dc.date.accessioned2023-09-26T08:50:07Z-
dc.date.available2023-09-26T08:50:07Z-
dc.date.issued2022-05-
dc.identifier.issn1225-4428-
dc.identifier.issn2287-3775-
dc.identifier.urihttps://scholarworks.bwise.kr/hanyang/handle/2021.sw.hanyang/191208-
dc.description.abstract본 논문에서는 기존 자연어 처리 분야에서 뛰어난 성능을 보이는 트랜스포머를 하이브리드 음성인식에서의음향모델로 사용하였다. 트랜스포머 음향모델은 attention 구조를 사용하여 시계열 데이터를 처리하며 연산량이 낮으면서 높은 성능을 보인다. 본 논문은 이러한 트랜스포머 AM에 기존 DNN-HMM 모델에서 사용하는 가중 유한 상태 전이기(weighted Finite-State Transducer, wFST) 기반 학습인 시퀀스 분류 학습의 네 가지 알고리즘을 각각 적용하여 성능을 높이는 방법을 제안한다. 또한 기존 Cross Entropy(CE)를 사용한 학습방식과 비교하여 5 %의 상대적word error rate(WER) 감소율을 보였다.-
dc.description.abstractIn this paper, we adopt a transformer that shows remarkable performance in natural language processing as an acoustic model of hybrid speech recognition. The transformer acoustic model uses attention structures to process sequential data and shows high performance with low computational cost. This paper proposes a method to improve the performance of transformer AM by applying each of the four algorithms of sequence discriminative training, a weighted finite-state transducer (wFST)-based learning used in the existing DNN-HMM model. In addition, compared to the Cross Entropy (CE) learning method, sequence discriminative method shows 5 % of the relative Word Error Rate (WER).-
dc.format.extent7-
dc.language한국어-
dc.language.isoKOR-
dc.publisher한국음향학회-
dc.titleSequence dicriminative training 기법을 사용한 트랜스포머 기반 음향 모델 성능 향상-
dc.title.alternativeImproving transformer-based acoustic model performance using sequence discriminative training-
dc.typeArticle-
dc.publisher.location대한민국-
dc.identifier.doi10.7776/ASK.2022.41.3.335-
dc.identifier.scopusid2-s2.0-85133242678-
dc.identifier.wosid000810472000009-
dc.identifier.bibliographicCitation한국음향학회지, v.41, no.3, pp 335 - 341-
dc.citation.title한국음향학회지-
dc.citation.volume41-
dc.citation.number3-
dc.citation.startPage335-
dc.citation.endPage341-
dc.type.docTypeArticle-
dc.identifier.kciidART002844665-
dc.description.isOpenAccessN-
dc.description.journalRegisteredClassscopus-
dc.description.journalRegisteredClassesci-
dc.description.journalRegisteredClasskci-
dc.relation.journalResearchAreaAcoustics-
dc.relation.journalWebOfScienceCategoryAcoustics-
dc.subject.keywordAuthorSpeech recognition-
dc.subject.keywordAuthorTransformer-
dc.subject.keywordAuthorSequence discriminative training-
dc.subject.keywordAuthorWeighted finite state transducer-
dc.subject.keywordAuthor음성인식-
dc.subject.keywordAuthor트랜스포머-
dc.subject.keywordAuthor시퀀스 분류 학습-
dc.subject.keywordAuthor가중 유한 상태 전이기-
dc.identifier.urlhttp://koreascience.or.kr/article/JAKO202216466662641.page-
Files in This Item
Appears in
Collections
서울 공과대학 > 서울 융합전자공학부 > 1. Journal Articles

qrcode

Items in ScholarWorks are protected by copyright, with all rights reserved, unless otherwise indicated.

Related Researcher

Researcher Chang, Joon-Hyuk photo

Chang, Joon-Hyuk
COLLEGE OF ENGINEERING (SCHOOL OF ELECTRONIC ENGINEERING)
Read more

Altmetrics

Total Views & Downloads

BROWSE