{"created":"2026-02-18T10:53:14.549369+00:00","metadata":{"_oai":{"id":"oai:ipsj.ixsq.nii.ac.jp:02007684","sets":["1164:5159:1771204180543:1771204240542"]},"path":["1771204240542"],"owner":"80578","recid":"2007684","title":["共有IPAエキスパートを持つswitch Conformerによる多言語音声認識"],"pubdate":{"attribute_name":"PubDate","attribute_value":"2026-02-24"},"_buckets":{"deposit":"7f673ff1-dca9-4236-b0ba-ea1079453f5d"},"_deposit":{"id":"2007684","pid":{"type":"depid","value":"2007684","revision_id":0},"owners":[80578],"status":"published","created_by":80578},"item_title":"共有IPAエキスパートを持つswitch Conformerによる多言語音声認識","author_link":[],"item_titles":{"attribute_name":"タイトル","attribute_value_mlt":[{"subitem_title":"共有IPAエキスパートを持つswitch Conformerによる多言語音声認識","subitem_title_language":"ja"}]},"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":"2026-02-24","item_4_text_3":{"attribute_name":"著者所属","attribute_value_mlt":[{"subitem_text_value":"NTT"},{"subitem_text_value":"NTT"},{"subitem_text_value":"京都大学情報学研究科"}]},"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/2007684/files/IPSJ-SLP26159115.pdf","label":"IPSJ-SLP26159115.pdf"},"date":[{"dateType":"Available","dateValue":"2028-02-24"}],"format":"application/pdf","billing":["billing_file"],"filename":"IPSJ-SLP26159115.pdf","filesize":[{"value":"1.1 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":"a1e8c8ef-ab8c-4a20-97b8-2498b408e30a","displaytype":"detail","licensetype":"license_note","license_note":"Copyright (c) 2026 by the Information Processing Society of Japan"}]},"item_4_creator_5":{"attribute_name":"著者名","attribute_type":"creator","attribute_value_mlt":[{"creatorNames":[{"creatorName":"三村,正人"}]},{"creatorNames":[{"creatorName":"Lee,Jaeyoung"}]},{"creatorNames":[{"creatorName":"河原,達也"}]}]},"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":"多言語対応のend-to-end音声認識（ASR）は、言語ごとに異なる文字体系、語彙、文法構造を扱う必要があるため、大きな課題を伴う。既存手法の多くは、言語間で十分な性能を得るために大規模モデルに依存しており、その結果、高い計算コストを要する。本研究では、この問題に対処するため、標準的なConformerとほぼ同等の推論コストを維持しつつ、モデル容量を拡張可能なSwitch Conformerを提案する。本手法では、各Conformerブロック内のFFNモジュールを、独立した複数のエキスパートからなる疎なmixtureに置き換え、入力ごとに1つのエキスパートのみを活性化することで、言語に依存した特徴を効率的にモデリングする。さらに、音素レベルの教師信号を用いて学習された共有エキスパートを導入し、言語に依存しない普遍的な音声特性を捉える。CommonVoiceデータセットを用いたストリーミングASRの実験により、これらのエキスパートが相互に補完的に機能し、追加される有効パラメータ数を最小限に抑えながら、ベースラインのConformerを上回る性能を達成することを示した。","subitem_description_type":"Other"}]},"item_4_biblio_info_10":{"attribute_name":"書誌情報","attribute_value_mlt":[{"bibliographicPageEnd":"7","bibliographic_titles":[{"bibliographic_title":"研究報告音声言語情報処理（SLP）"}],"bibliographicPageStart":"1","bibliographicIssueDates":{"bibliographicIssueDate":"2026-02-24","bibliographicIssueDateType":"Issued"},"bibliographicIssueNumber":"115","bibliographicVolumeNumber":"2026-SLP-159"}]},"relation_version_is_last":true,"weko_creator_id":"80578"},"id":2007684,"updated":"2026-02-18T10:53:19.701597+00:00","links":{}}