{"links":{},"metadata":{"_oai":{"id":"oai:ipsj.ixsq.nii.ac.jp:00202993","sets":["1164:5159:10092:10136"]},"path":["10136"],"owner":"44499","recid":"202993","title":["ストリーミング注意機構型sequence-to-sequenceモデルによる講演音声認識"],"pubdate":{"attribute_name":"公開日","attribute_value":"2020-02-06"},"_buckets":{"deposit":"03a58fb4-2b15-4f07-982c-0c47b6fd6ab3"},"_deposit":{"id":"202993","pid":{"type":"depid","value":"202993","revision_id":0},"owners":[44499],"status":"published","created_by":44499},"item_title":"ストリーミング注意機構型sequence-to-sequenceモデルによる講演音声認識","author_link":["499224","499225","499226"],"item_titles":{"attribute_name":"タイトル","attribute_value_mlt":[{"subitem_title":"ストリーミング注意機構型sequence-to-sequenceモデルによる講演音声認識"}]},"item_keyword":{"attribute_name":"キーワード","attribute_value_mlt":[{"subitem_subject":"一般講演2","subitem_subject_scheme":"Other"}]},"item_type_id":"4","publish_date":"2020-02-06","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"}]},"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/202993/files/IPSJ-SLP20131009.pdf","label":"IPSJ-SLP20131009.pdf"},"date":[{"dateType":"Available","dateValue":"2022-02-06"}],"format":"application/pdf","billing":["billing_file"],"filename":"IPSJ-SLP20131009.pdf","filesize":[{"value":"4.2 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":"aee6dceb-30c3-4824-a8b6-5c030a36613d","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":"稲熊, 寛文"}],"nameIdentifiers":[{}]},{"creatorNames":[{"creatorName":"三村, 正人"}],"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":"ラベル同期型モデルである注意機構型 sequence-to-sequence では，入力音声フレームごとの出力を獲得できないためストリーミング認識には不向きであった．これに対処するため，逐次トークンを出力できるモデルがこれまで提案されてきたが，アライメント境界は過去のアライメントに依存するため，出力系列長が長くなるにつれて認識誤りの影響が後方のトークン生成へ伝搬するという問題があった．そこで本稿では，CTC のアライメント情報を教師として学習中に与えることにより，アライメントの学習を頑健にする CTC 同期学習を提案する．また，一定区間の入力フレームごとに枝刈りを行うチャンク同期型ビームサーチを提案し，外部の音声区間検出モデルを用いない，講演単位の認識を行う．日本語話し言葉コーパスにおける実験において，CTC 同期学習による認識精度の改善を確認する．","subitem_description_type":"Other"}]},"item_4_biblio_info_10":{"attribute_name":"書誌情報","attribute_value_mlt":[{"bibliographicPageEnd":"7","bibliographic_titles":[{"bibliographic_title":"研究報告音声言語情報処理（SLP）"}],"bibliographicPageStart":"1","bibliographicIssueDates":{"bibliographicIssueDate":"2020-02-06","bibliographicIssueDateType":"Issued"},"bibliographicIssueNumber":"9","bibliographicVolumeNumber":"2020-SLP-131"}]},"relation_version_is_last":true,"weko_creator_id":"44499"},"updated":"2025-01-19T20:41:53.780606+00:00","created":"2025-01-19T01:05:29.098562+00:00","id":202993}