{"id":216616,"updated":"2025-01-19T15:47:34.674803+00:00","links":{},"created":"2025-01-19T01:17:08.413888+00:00","metadata":{"_oai":{"id":"oai:ipsj.ixsq.nii.ac.jp:00216616","sets":["1164:5159:10869:10870"]},"path":["10870"],"owner":"44499","recid":"216616","title":["End-to-End方言音声認識のための方言ラベルを考慮した半教師あり学習"],"pubdate":{"attribute_name":"公開日","attribute_value":"2022-02-22"},"_buckets":{"deposit":"97adb753-0ef0-46d9-b6f4-d7c64787f89f"},"_deposit":{"id":"216616","pid":{"type":"depid","value":"216616","revision_id":0},"owners":[44499],"status":"published","created_by":44499},"item_title":"End-to-End方言音声認識のための方言ラベルを考慮した半教師あり学習","author_link":["559290","559287","559288","559289"],"item_titles":{"attribute_name":"タイトル","attribute_value_mlt":[{"subitem_title":"End-to-End方言音声認識のための方言ラベルを考慮した半教師あり学習"}]},"item_keyword":{"attribute_name":"キーワード","attribute_value_mlt":[{"subitem_subject":"SLP","subitem_subject_scheme":"Other"}]},"item_type_id":"4","publish_date":"2022-02-22","item_4_text_3":{"attribute_name":"著者所属","attribute_value_mlt":[{"subitem_text_value":"現在, 東京都立大学システムデザイン研究科情報科学域"},{"subitem_text_value":"日本電信電話株式会社"},{"subitem_text_value":"現在, 東京都立大学システムデザイン研究科情報科学域"},{"subitem_text_value":"現在, 東京都立大学システムデザイン研究科情報科学域"}]},"item_4_text_4":{"attribute_name":"著者所属(英)","attribute_value_mlt":[{"subitem_text_value":"Presently with Tokyo Metropolitan University, Faculty School of Systems Design, Department of Computer Science","subitem_text_language":"en"},{"subitem_text_value":"Nippon Telegraph and Telephone","subitem_text_language":"en"},{"subitem_text_value":"Presently with Tokyo Metropolitan University, Faculty School of Systems Design, Department of Computer Science","subitem_text_language":"en"},{"subitem_text_value":"Presently with Tokyo Metropolitan University, Faculty School of Systems Design, Department of Computer Science","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/216616/files/IPSJ-SLP22140015.pdf","label":"IPSJ-SLP22140015.pdf"},"date":[{"dateType":"Available","dateValue":"2024-02-22"}],"format":"application/pdf","billing":["billing_file"],"filename":"IPSJ-SLP22140015.pdf","filesize":[{"value":"1.2 MB"}],"mimetype":"application/pdf","priceinfo":[{"tax":["include_tax"],"price":"660","billingrole":"5"},{"tax":["include_tax"],"price":"330","billingrole":"6"},{"tax":["include_tax"],"price":"0","billingrole":"22"},{"tax":["include_tax"],"price":"0","billingrole":"44"}],"accessrole":"open_date","version_id":"fe1a5ae5-02ec-41ab-83db-959fa5aa986b","displaytype":"detail","licensetype":"license_note","license_note":"Copyright (c) 2022 by the Information Processing Society of Japan"}]},"item_4_creator_5":{"attribute_name":"著者名","attribute_type":"creator","attribute_value_mlt":[{"creatorNames":[{"creatorName":"今泉, 遼"}],"nameIdentifiers":[{}]},{"creatorNames":[{"creatorName":"増村, 亮"}],"nameIdentifiers":[{}]},{"creatorNames":[{"creatorName":"塩田, さやか"}],"nameIdentifiers":[{}]},{"creatorNames":[{"creatorName":"貴家, 仁志"}],"nameIdentifiers":[{}]}]},"item_4_source_id_9":{"attribute_name":"書誌レコードID","attribute_value_mlt":[{"subitem_source_identifier":"AN10442647","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":"2188-8663","subitem_source_identifier_type":"ISSN"}]},"item_4_description_7":{"attribute_name":"論文抄録","attribute_value_mlt":[{"subitem_description":"本論文では，方言に対してロバストな End-to-End 音声認識のための方言ラベルを考慮した半教師あり学習を提案する．最先端の深層学習に基づく音声認識手法である End-to-End 音声認識モデルは学習に大量の音声と書き起こしのペアデータが必要となることが知られている．また，これまでに音声と書き起こしのペアデータのうち書き起こしだけがない音声データが大量にある場合，音声データをモデル学習に有効活用する方法として半教師あり学習が提案されている．半教師あり学習法の 1 つとして，ペアデータが揃っている小規模な学習データを用いて教師モデルを学習し，教師モデルを用いて生成した自動書き起こしをペアデータとして補完することで大規模なデータの活用を可能としている．日本語方言のための方言ラベルを考慮した End-to-End 音声認識モデルでは書き起こしだけでなく方言ラベルを用いてモデル学習を行なっている．しかし，音声データに対して方言ラベルと書き起こしのペアデータが揃っているデータの収集は困難である．そこで本研究では，方言ラベルのみ付与された音声データが大量にある場合を考え，方言音声データの活用のために方言ラベルを考慮した音声認識モデルのために半教師学習を用いることを提案する．提案法では，方言ラベルを考慮することで方言の書き起こしの精度を向上することが可能となるため，最終的な半教師あり学習によって得られるモデルが方言に対して頑健になることが期待できる．実験において提案法は方言ラベルを考慮しない半教師あり学習よりも CER が改善することを報告する．","subitem_description_type":"Other"}]},"item_4_biblio_info_10":{"attribute_name":"書誌情報","attribute_value_mlt":[{"bibliographicPageEnd":"6","bibliographic_titles":[{"bibliographic_title":"研究報告音声言語情報処理（SLP）"}],"bibliographicPageStart":"1","bibliographicIssueDates":{"bibliographicIssueDate":"2022-02-22","bibliographicIssueDateType":"Issued"},"bibliographicIssueNumber":"15","bibliographicVolumeNumber":"2022-SLP-140"}]},"relation_version_is_last":true,"weko_creator_id":"44499"}}