{"metadata":{"_oai":{"id":"oai:ipsj.ixsq.nii.ac.jp:00214082","sets":["1164:4179:10535:10759"]},"path":["10759"],"owner":"44499","recid":"214082","title":["End-to-End音声認識のための粒度の異なるサブワード単位に基づく階層的な条件づけ"],"pubdate":{"attribute_name":"公開日","attribute_value":"2021-11-24"},"_buckets":{"deposit":"1339a02a-ff31-4b4f-8385-7d3b89f6ab14"},"_deposit":{"id":"214082","pid":{"type":"depid","value":"214082","revision_id":0},"owners":[44499],"status":"published","created_by":44499},"item_title":"End-to-End音声認識のための粒度の異なるサブワード単位に基づく階層的な条件づけ","author_link":["548563","548557","548562","548559","548560","548561","548556","548558"],"item_titles":{"attribute_name":"タイトル","attribute_value_mlt":[{"subitem_title":"End-to-End音声認識のための粒度の異なるサブワード単位に基づく階層的な条件づけ"},{"subitem_title":"End-to-End Speech Recognition with Multi-Granular Subword Units and Hierarchical Conditioning Mechanism","subitem_title_language":"en"}]},"item_keyword":{"attribute_name":"キーワード","attribute_value_mlt":[{"subitem_subject":"深層学習モデル","subitem_subject_scheme":"Other"}]},"item_type_id":"4","publish_date":"2021-11-24","item_4_text_3":{"attribute_name":"著者所属","attribute_value_mlt":[{"subitem_text_value":"早稲田大学"},{"subitem_text_value":"早稲田大学"},{"subitem_text_value":"早稲田大学"},{"subitem_text_value":"早稲田大学"}]},"item_4_text_4":{"attribute_name":"著者所属(英)","attribute_value_mlt":[{"subitem_text_value":"Waseda University","subitem_text_language":"en"},{"subitem_text_value":"Waseda University","subitem_text_language":"en"},{"subitem_text_value":"Waseda University","subitem_text_language":"en"},{"subitem_text_value":"Waseda University","subitem_text_language":"en"}]},"item_language":{"attribute_name":"言語","attribute_value_mlt":[{"subitem_language":"jpn"}]},"item_publisher":{"attribute_name":"出版者","attribute_value_mlt":[{"subitem_publisher":"情報処理学会","subitem_publisher_language":"ja"}]},"publish_status":"0","weko_shared_id":-1,"item_file_price":{"attribute_name":"Billing file","attribute_type":"file","attribute_value_mlt":[{"url":{"url":"https://ipsj.ixsq.nii.ac.jp/record/214082/files/IPSJ-NL21251019.pdf","label":"IPSJ-NL21251019.pdf"},"date":[{"dateType":"Available","dateValue":"2023-11-24"}],"format":"application/pdf","billing":["billing_file"],"filename":"IPSJ-NL21251019.pdf","filesize":[{"value":"380.9 kB"}],"mimetype":"application/pdf","priceinfo":[{"tax":["include_tax"],"price":"660","billingrole":"5"},{"tax":["include_tax"],"price":"330","billingrole":"6"},{"tax":["include_tax"],"price":"0","billingrole":"23"},{"tax":["include_tax"],"price":"0","billingrole":"44"}],"accessrole":"open_date","version_id":"cddc4bda-8e08-4250-bca7-d642aafead69","displaytype":"detail","licensetype":"license_note","license_note":"Copyright (c) 2021 by the Information Processing Society of Japan"}]},"item_4_creator_5":{"attribute_name":"著者名","attribute_type":"creator","attribute_value_mlt":[{"creatorNames":[{"creatorName":"樋口, 陽祐"}],"nameIdentifiers":[{}]},{"creatorNames":[{"creatorName":"軽部, 敬太"}],"nameIdentifiers":[{}]},{"creatorNames":[{"creatorName":"小川, 哲司"}],"nameIdentifiers":[{}]},{"creatorNames":[{"creatorName":"小林, 哲則"}],"nameIdentifiers":[{}]}]},"item_4_creator_6":{"attribute_name":"著者名(英)","attribute_type":"creator","attribute_value_mlt":[{"creatorNames":[{"creatorName":"Yosuke, Higuchi","creatorNameLang":"en"}],"nameIdentifiers":[{}]},{"creatorNames":[{"creatorName":"Keita, Karube","creatorNameLang":"en"}],"nameIdentifiers":[{}]},{"creatorNames":[{"creatorName":"Tetsuji, Ogawa","creatorNameLang":"en"}],"nameIdentifiers":[{}]},{"creatorNames":[{"creatorName":"Tetsunori, Kobayashi","creatorNameLang":"en"}],"nameIdentifiers":[{}]}]},"item_4_source_id_9":{"attribute_name":"書誌レコードID","attribute_value_mlt":[{"subitem_source_identifier":"AN10115061","subitem_source_identifier_type":"NCID"}]},"item_4_textarea_12":{"attribute_name":"Notice","attribute_value_mlt":[{"subitem_textarea_value":"SIG Technical Reports are nonrefereed and hence may later appear in any journals, conferences, symposia, etc."}]},"item_resource_type":{"attribute_name":"資源タイプ","attribute_value_mlt":[{"resourceuri":"http://purl.org/coar/resource_type/c_18gh","resourcetype":"technical report"}]},"item_4_source_id_11":{"attribute_name":"ISSN","attribute_value_mlt":[{"subitem_source_identifier":"2188-8779","subitem_source_identifier_type":"ISSN"}]},"item_4_description_7":{"attribute_name":"論文抄録","attribute_value_mlt":[{"subitem_description":"End-to-End 音声認識では，単語を推定するのに適した特徴表現が暗黙的に獲得されることを期待している．しかし，入力の音声信号と出力の単語列では，情報の抽象度が大きく異なるため，目的の特徴表現を End-to-End に学習するのは困難である．本研究では，End-to-End 音声認識において，単語単位の特徴表現を効果的に学習するために，Connectionist Temporal Classification (CTC) に基づいた階層的条件付きモデルを提案する．提案モデルでは，最終層に加えて，複数の中間層に対して CTC 損失を適用し，各 CTC における出力単位の粒度を最終層に向けて段階的に高くする．このとき，粒度の低い単位による予測によって粒度の高い単位による予測を条件付けることで，単語単位の系列に対する生成過程を明示的に学習する．言語情報の抽象度が段階的に組み上がるモデルを構築することで，単語単位の特徴表現が効果的に学習されることを期待する．LibriSpeech-{100h, 960h} と TEDLIUM2 を用いた実験において提案モデルを評価したところ，既存モデルよりも高い認識性能を与えることが明らかとなった．また，詳細な分析の結果，提案モデルによって単語単位の認識に適した特徴表現が学習できることを確認した．","subitem_description_type":"Other"}]},"item_4_biblio_info_10":{"attribute_name":"書誌情報","attribute_value_mlt":[{"bibliographicPageEnd":"8","bibliographic_titles":[{"bibliographic_title":"研究報告自然言語処理（NL）"}],"bibliographicPageStart":"1","bibliographicIssueDates":{"bibliographicIssueDate":"2021-11-24","bibliographicIssueDateType":"Issued"},"bibliographicIssueNumber":"19","bibliographicVolumeNumber":"2021-NL-251"}]},"relation_version_is_last":true,"weko_creator_id":"44499"},"id":214082,"updated":"2025-01-19T16:54:33.151935+00:00","links":{},"created":"2025-01-19T01:14:55.046455+00:00"}