{"metadata":{"_oai":{"id":"oai:ipsj.ixsq.nii.ac.jp:00232524","sets":["1164:5159:11541:11549"]},"path":["11549"],"owner":"44499","recid":"232524","title":["多言語SSLモデルに基づく音声認識のノンパラレルコーパスのみを用いたドメイン適応"],"pubdate":{"attribute_name":"公開日","attribute_value":"2024-02-22"},"_buckets":{"deposit":"a1da5ec4-9cf5-4871-b266-8a1611f843a7"},"_deposit":{"id":"232524","pid":{"type":"depid","value":"232524","revision_id":0},"owners":[44499],"status":"published","created_by":44499},"item_title":"多言語SSLモデルに基づく音声認識のノンパラレルコーパスのみを用いたドメイン適応","author_link":["629552","629547","629550","629548","629551","629555","629556","629549","629553","629554"],"item_titles":{"attribute_name":"タイトル","attribute_value_mlt":[{"subitem_title":"多言語SSLモデルに基づく音声認識のノンパラレルコーパスのみを用いたドメイン適応"},{"subitem_title":"Domain adaptation of speech recognition model based on multilingual SSL model using only nonparallel corpus.","subitem_title_language":"en"}]},"item_keyword":{"attribute_name":"キーワード","attribute_value_mlt":[{"subitem_subject":"ポスターセッション2 SP/SLP","subitem_subject_scheme":"Other"}]},"item_type_id":"4","publish_date":"2024-02-22","item_4_text_3":{"attribute_name":"著者所属","attribute_value_mlt":[{"subitem_text_value":"豊橋技術科学大学"},{"subitem_text_value":"日本電信電話株式会社"},{"subitem_text_value":"豊橋技術科学大学"},{"subitem_text_value":"阿南工業高等専門学校"},{"subitem_text_value":"豊橋技術科学大学"}]},"item_4_text_4":{"attribute_name":"著者所属(英)","attribute_value_mlt":[{"subitem_text_value":"Toyohashi University of Technology","subitem_text_language":"en"},{"subitem_text_value":"NIPPON TELEGRAPH AND TELEPHONE CORPORATION","subitem_text_language":"en"},{"subitem_text_value":"Toyohashi University of Technology","subitem_text_language":"en"},{"subitem_text_value":"National Institute of Technology, Anan College","subitem_text_language":"en"},{"subitem_text_value":"Toyohashi University of Technology","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/232524/files/IPSJ-SLP24151054.pdf","label":"IPSJ-SLP24151054.pdf"},"format":"application/pdf","billing":["billing_file"],"filename":"IPSJ-SLP24151054.pdf","filesize":[{"value":"941.4 kB"}],"mimetype":"application/pdf","priceinfo":[{"tax":["include_tax"],"price":"0","billingrole":"22"},{"tax":["include_tax"],"price":"0","billingrole":"44"}],"accessrole":"open_login","version_id":"c2c4065e-d455-497d-81ca-8ee110dc4820","displaytype":"detail","licensetype":"license_note","license_note":"Copyright (c) 2024 by the Institute of Electronics, Information and Communication Engineers This SIG report is only available to those in membership of the SIG."}]},"item_4_creator_5":{"attribute_name":"著者名","attribute_type":"creator","attribute_value_mlt":[{"creatorNames":[{"creatorName":"木内, 貴浩"}],"nameIdentifiers":[{}]},{"creatorNames":[{"creatorName":"小川, 厚徳"}],"nameIdentifiers":[{}]},{"creatorNames":[{"creatorName":"若林, 佑幸"}],"nameIdentifiers":[{}]},{"creatorNames":[{"creatorName":"太田, 健吾"}],"nameIdentifiers":[{}]},{"creatorNames":[{"creatorName":"北岡, 教英"}],"nameIdentifiers":[{}]}]},"item_4_creator_6":{"attribute_name":"著者名(英)","attribute_type":"creator","attribute_value_mlt":[{"creatorNames":[{"creatorName":"Takahiro, Kinouchi","creatorNameLang":"en"}],"nameIdentifiers":[{}]},{"creatorNames":[{"creatorName":"Atsunori, Ogawa","creatorNameLang":"en"}],"nameIdentifiers":[{}]},{"creatorNames":[{"creatorName":"Yukoh, Wakabayashi","creatorNameLang":"en"}],"nameIdentifiers":[{}]},{"creatorNames":[{"creatorName":"Kengo, Ohta","creatorNameLang":"en"}],"nameIdentifiers":[{}]},{"creatorNames":[{"creatorName":"Norihide, Kitaoka","creatorNameLang":"en"}],"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":"音声認識（automatic speech recognition:  ASR）は様々なサービスや事業で活用されており，それらの各ドメインの音声の認識精度は，そのドメインにおける音声データ量に依存することが知られている．一般的には，目標ドメインの大量の音声データとその書き起こしテキストデータで ASR モデルを学習することが望ましいが，音声とその書き起こしテキストのペアデータを構築するのは非常に高コストである．そこで我々は，音声のみ，もしくはテキストのみであれば比較的低コストで収集可能であることに着目する．この条件のもと，本研究では自己教師あり学習 (self-supervised learning: SSL) モデルを基にした音声のみを用いたドメイン適応と，言語モデル統合によるテキストのみを用いたドメイン適応の２つのアプローチによって音声認識モデルを目標ドメインへ適応することを実現する．実験の結果，英語音声で事前学習された SSL モデルを用いた場合，9.2 pts のCER の低下が見られ，多言語音声で事前学習された SSL モデルを用いた場合，5.3 pts の CER の低下が見られ，精度が向上した．","subitem_description_type":"Other"}]},"item_4_description_8":{"attribute_name":"論文抄録(英)","attribute_value_mlt":[{"subitem_description":"Automatic speech recognition (ASR) models are used in various services and businesses, and each domain’s recognition accuracy depends on the amount of speech data in that domain. Generally, it is desirable to train an ASR model with a large amount of target-domain speech data and its transcribed text data. However, it is high cost to construct these data. Therefore, we focus on collecting only speech or text data with low cost. Under these conditions, we use two approaches: domain adaptation using only speech data based on a self-supervised learning (SSL) model and domain adaptation using only text data based on language model fusion. In the experiment, the CER improved by 9.2 pts when using the SSL model pre-trained on English speech and by 5.3 pts when using the SSL model pre-trained on multilingual speech.","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":"2024-02-22","bibliographicIssueDateType":"Issued"},"bibliographicIssueNumber":"54","bibliographicVolumeNumber":"2024-SLP-151"}]},"relation_version_is_last":true,"weko_creator_id":"44499"},"id":232524,"updated":"2025-01-19T10:25:10.048691+00:00","links":{},"created":"2025-01-19T01:33:26.533490+00:00"}