{"metadata":{"_oai":{"id":"oai:ipsj.ixsq.nii.ac.jp:00241576","sets":["1164:4179:11560:11869"]},"path":["11869"],"owner":"44499","recid":"241576","title":["再帰的フィードバックを用いた階層的End-to-End音声認識"],"pubdate":{"attribute_name":"公開日","attribute_value":"2024-12-05"},"_buckets":{"deposit":"491e08c5-90cb-4b9d-8df1-2c1d8afe4f61"},"_deposit":{"id":"241576","pid":{"type":"depid","value":"241576","revision_id":0},"owners":[44499],"status":"published","created_by":44499},"item_title":"再帰的フィードバックを用いた階層的End-to-End音声認識","author_link":["665327","665324","665326","665328","665329","665325","665322","665323"],"item_titles":{"attribute_name":"タイトル","attribute_value_mlt":[{"subitem_title":"再帰的フィードバックを用いた階層的End-to-End音声認識"},{"subitem_title":"Hierarchical End-to-End Speech Recognition with Recursive Feedback","subitem_title_language":"en"}]},"item_keyword":{"attribute_name":"キーワード","attribute_value_mlt":[{"subitem_subject":"音声認識","subitem_subject_scheme":"Other"}]},"item_type_id":"4","publish_date":"2024-12-05","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/241576/files/IPSJ-NL24262001.pdf","label":"IPSJ-NL24262001.pdf"},"date":[{"dateType":"Available","dateValue":"2026-12-05"}],"format":"application/pdf","billing":["billing_file"],"filename":"IPSJ-NL24262001.pdf","filesize":[{"value":"467.8 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":"e8771569-ca09-4dff-9567-be3be0c8f31e","displaytype":"detail","licensetype":"license_note","license_note":"Copyright (c) 2024 by the Information Processing Society of Japan"}]},"item_4_creator_5":{"attribute_name":"著者名","attribute_type":"creator","attribute_value_mlt":[{"creatorNames":[{"creatorName":"楠, 奈穂美"}],"nameIdentifiers":[{}]},{"creatorNames":[{"creatorName":"樋口, 陽祐"}],"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":"Nahomi, Kusunoki","creatorNameLang":"en"}],"nameIdentifiers":[{}]},{"creatorNames":[{"creatorName":"Yosuke, Higuchi","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，単一のニューラルネットワークにより細粒度の音声信号から粗粒度の言語記号列への変換を行う．従来の End-to-End モデルでは入出力間の抽象度の差が，記号列の認識に適した特徴表現の学習や音声認識の精度向上を妨げる要因となっている．これに対し，階層的マルチタスク学習モデルでは中間層に補助的な損失を導入し，出力単位の粒度を徐々に高めることにより，粗粒度の系列推定に適した中間表現を学習することが可能になる．本研究では，階層的マルチタスク学習モデルにおける階層構造を強化するため，Connectionist Temporal Classification（CTC）に基づいた，再帰的フィードバックを用いた階層的音声認識モデルを提案する．提案モデルでは同一のモデル層を再帰的に利用し中間予測を洗練する．予測を同層での入力に明示的に条件付けることにより中間表現が洗練され，より高層での正確な予測を可能にする．LibriSpeech，TEDLIUM2，日本語話し言葉コーパスを用いた実験により提案モデルを評価したところ，提案モデルは既存の階層的マルチタスク学習モデルの認識性能を上回ることが明らかになった．また，詳細な分析により，提案モデルは単一のモデル内で精度と推論速度のバランスを制御する能力を有することが示された．","subitem_description_type":"Other"}]},"item_4_biblio_info_10":{"attribute_name":"書誌情報","attribute_value_mlt":[{"bibliographicPageEnd":"7","bibliographic_titles":[{"bibliographic_title":"研究報告自然言語処理（NL）"}],"bibliographicPageStart":"1","bibliographicIssueDates":{"bibliographicIssueDate":"2024-12-05","bibliographicIssueDateType":"Issued"},"bibliographicIssueNumber":"1","bibliographicVolumeNumber":"2024-NL-262"}]},"relation_version_is_last":true,"weko_creator_id":"44499"},"id":241576,"updated":"2025-01-19T07:36:57.688000+00:00","links":{},"created":"2025-01-19T01:46:15.762826+00:00"}