{"links":{},"metadata":{"_oai":{"id":"oai:ipsj.ixsq.nii.ac.jp:00208151","sets":["1164:5159:10092:10413"]},"path":["10413"],"owner":"44499","recid":"208151","title":["BERTによるSequence-to-Sequence音声認識への知識蒸留"],"pubdate":{"attribute_name":"公開日","attribute_value":"2020-11-25"},"_buckets":{"deposit":"89b6d20d-a990-45d7-99f8-0e390a22311f"},"_deposit":{"id":"208151","pid":{"type":"depid","value":"208151","revision_id":0},"owners":[44499],"status":"published","created_by":44499},"item_title":"BERTによるSequence-to-Sequence音声認識への知識蒸留","author_link":["520924","520927","520923","520922","520925","520926"],"item_titles":{"attribute_name":"タイトル","attribute_value_mlt":[{"subitem_title":"BERTによるSequence-to-Sequence音声認識への知識蒸留"}]},"item_keyword":{"attribute_name":"キーワード","attribute_value_mlt":[{"subitem_subject":"音声検索・認識","subitem_subject_scheme":"Other"}]},"item_type_id":"4","publish_date":"2020-11-25","item_4_text_3":{"attribute_name":"著者所属","attribute_value_mlt":[{"subitem_text_value":"京都大学大学院情報学研究科"},{"subitem_text_value":"京都大学大学院情報学研究科"},{"subitem_text_value":"京都大学大学院情報学研究科"},{"subitem_text_value":"京都大学大学院情報学研究科"},{"subitem_text_value":"京都大学大学院情報学研究科"},{"subitem_text_value":"京都大学大学院情報学研究科"}]},"item_4_text_4":{"attribute_name":"著者所属(英)","attribute_value_mlt":[{"subitem_text_value":"Graduate School of Informatics, Kyoto University","subitem_text_language":"en"},{"subitem_text_value":"Graduate School of Informatics, Kyoto University","subitem_text_language":"en"},{"subitem_text_value":"Graduate School of Informatics, Kyoto University","subitem_text_language":"en"},{"subitem_text_value":"Graduate School of Informatics, Kyoto University","subitem_text_language":"en"},{"subitem_text_value":"Graduate School of Informatics, Kyoto University","subitem_text_language":"en"},{"subitem_text_value":"Graduate School of Informatics, Kyoto 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/208151/files/IPSJ-SLP20134002.pdf","label":"IPSJ-SLP20134002.pdf"},"date":[{"dateType":"Available","dateValue":"2022-11-25"}],"format":"application/pdf","billing":["billing_file"],"filename":"IPSJ-SLP20134002.pdf","filesize":[{"value":"2.4 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":"f933ddfc-9e29-4924-ad52-76ecfc165ce6","displaytype":"detail","licensetype":"license_note","license_note":"Copyright (c) 2020 by the Information Processing Society of Japan"}]},"item_4_creator_5":{"attribute_name":"著者名","attribute_type":"creator","attribute_value_mlt":[{"creatorNames":[{"creatorName":"二見, 颯"}],"nameIdentifiers":[{}]},{"creatorNames":[{"creatorName":"稲熊, 寛文"}],"nameIdentifiers":[{}]},{"creatorNames":[{"creatorName":"上乃, 聖"}],"nameIdentifiers":[{}]},{"creatorNames":[{"creatorName":"三村, 正人"}],"nameIdentifiers":[{}]},{"creatorNames":[{"creatorName":"坂井, 信輔"}],"nameIdentifiers":[{}]},{"creatorNames":[{"creatorName":"河原, 達也"}],"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":"近年，音声認識において Sequence-to-Sequence（Seq2Seq）モデルが注目されている．Seq2Seq 音声認識モデルは音声とテキストのペアデータから学習されるため，テキストデータを追加利用することが課題となっている．また，Seq2Seq モデルではある単語を予測するためにそれ以前の文脈が用いられ，以後の文脈を用いることができない．そこで，本研究では BERT をある単語の以前だけでなく以後の文脈を用いて予測を行う言語モデルとして Seq2Seq 音声認識へ適用する．適用法として BERT を教師モデル，Seq2Seq 音声認識モデルを生徒モデルとした知識蒸留法を提案する．ここで，BERT の入力として複数発話にまたがるコンテキストを利用する．主に日本語話し言葉コーパス（CSJ）上の評価実験によって，提案法による認識精度の大きな改善を確認した．さらに，従来の言語モデル適用法であるリスコアリングや Shallow Fusion と比較し，提案法は推論速度，認識精度ともに上回ることを確認した．","subitem_description_type":"Other"}]},"item_4_biblio_info_10":{"attribute_name":"書誌情報","attribute_value_mlt":[{"bibliographicPageEnd":"6","bibliographic_titles":[{"bibliographic_title":"研究報告音声言語情報処理（SLP）"}],"bibliographicPageStart":"1","bibliographicIssueDates":{"bibliographicIssueDate":"2020-11-25","bibliographicIssueDateType":"Issued"},"bibliographicIssueNumber":"2","bibliographicVolumeNumber":"2020-SLP-134"}]},"relation_version_is_last":true,"weko_creator_id":"44499"},"updated":"2025-01-19T18:54:28.874959+00:00","created":"2025-01-19T01:09:44.099654+00:00","id":208151}