{"metadata":{"_oai":{"id":"oai:ipsj.ixsq.nii.ac.jp:00050237","sets":["1164:4402:4418:4421"]},"path":["4421"],"owner":"1","recid":"50237","title":["音素片のカーネル主成分分析を用いたトピックセグメンテーション"],"pubdate":{"attribute_name":"公開日","attribute_value":"2005-03-14"},"_buckets":{"deposit":"88686443-5f9a-4b5d-a66d-c54c459d2cba"},"_deposit":{"id":"50237","pid":{"type":"depid","value":"50237","revision_id":0},"owners":[1],"status":"published","created_by":1},"item_title":"音素片のカーネル主成分分析を用いたトピックセグメンテーション","author_link":["0","0"],"item_titles":{"attribute_name":"タイトル","attribute_value_mlt":[{"subitem_title":"音素片のカーネル主成分分析を用いたトピックセグメンテーション"},{"subitem_title":"Topic Segmentation Using Kernel Pricipal Component Analysis for Sub-Phonetic Segments","subitem_title_language":"en"}]},"item_type_id":"4","publish_date":"2005-03-14","item_4_text_3":{"attribute_name":"著者所属","attribute_value_mlt":[{"subitem_text_value":"産業技術総合研究所"},{"subitem_text_value":"産業技術総合研究所"},{"subitem_text_value":"産業技術総合研究所"}]},"item_4_text_4":{"attribute_name":"著者所属(英)","attribute_value_mlt":[{"subitem_text_value":"National Institute of Advanced Industrial Science and Technology","subitem_text_language":"en"},{"subitem_text_value":"National Institute of Advanced Industrial Science and Technology","subitem_text_language":"en"},{"subitem_text_value":"National Institute of Advanced Industrial Science and 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/50237/files/IPSJ-ICS04139007.pdf"},"date":[{"dateType":"Available","dateValue":"2007-03-14"}],"format":"application/pdf","billing":["billing_file"],"filename":"IPSJ-ICS04139007.pdf","filesize":[{"value":"743.4 kB"}],"mimetype":"application/pdf","priceinfo":[{"tax":["include_tax"],"price":"660","billingrole":"5"},{"tax":["include_tax"],"price":"330","billingrole":"6"},{"tax":["include_tax"],"price":"0","billingrole":"25"},{"tax":["include_tax"],"price":"0","billingrole":"44"}],"accessrole":"open_date","version_id":"4e435f81-d7d0-4862-8bb7-bab6251c7cf2","displaytype":"detail","licensetype":"license_note","license_note":"Copyright (c) 2005 by the Information Processing Society of Japan"}]},"item_4_creator_5":{"attribute_name":"著者名","attribute_type":"creator","attribute_value_mlt":[{"creatorNames":[{"creatorName":"佐土原, 健"},{"creatorName":"李時旭"},{"creatorName":"児島, 宏明"}],"nameIdentifiers":[{}]}]},"item_4_creator_6":{"attribute_name":"著者名(英)","attribute_type":"creator","attribute_value_mlt":[{"creatorNames":[{"creatorName":"Ken, Sadohara","creatorNameLang":"en"},{"creatorName":"Shi-wook, Lee","creatorNameLang":"en"},{"creatorName":"Hiroaki, Kojima","creatorNameLang":"en"}],"nameIdentifiers":[{}]}]},"item_4_source_id_9":{"attribute_name":"書誌レコードID","attribute_value_mlt":[{"subitem_source_identifier":"AA11135936","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_description_7":{"attribute_name":"論文抄録","attribute_value_mlt":[{"subitem_description":"本論文では、語彙制約を用いることなしに、入力音声を意味的に等質な部分に分割する手法を提案する。この手法は、大語彙連続音声認識システム等によって、キーワードを抽出することなく、音声を、音素よりも粒度の細かい音素片の列として認識した上で、直接トピックセグメンテーションを行う。これにより、一定長以下の任意の音素片列に基づいた、語彙と文法に制約されないトピックセグメンテーションが可能になる。また、カーネル主成分分析を用いて、一つのトピックにおいて共起する音素片列を、まとめて一つの基底とすることによって、各分析区間を表現することも本手法の特徴である。これにより、ベクトルの余弦が、トピックに関する類似性を反映することになり、この余弦を類似性の指標として用いる階層的クラスタリング法により、トピック単位のクラスタリングを行う。また、このような手法の有用性を、ニュース音声のトピックセグメンテーションの実験により示す。","subitem_description_type":"Other"}]},"item_4_description_8":{"attribute_name":"論文抄録(英)","attribute_value_mlt":[{"subitem_description":"This paper describes an open-vocabulary method for segmenting spoken documents into topically homogeneous blocks. Without transcribing the spoken documents into texts, the method builds the topical clusters directly from recognized sub-phonetic segments, and thus it is not constrained in term of vocabulary or grammar. Each analysis interval constituting the clusters is represented as a vector in a high dimensional space spanned by all sub-phonetic segments with given length. Then a kernel principal component analysis reduces the dimensionality by grouping co-occurred sub-phonetic segments in each topic. This yields that cosine similarity between vectors is related with topical similarity, and the hierarchical clustering method using the similarity measure is expected to form topically homogeneous clusters. In fact,effectiveness of the method is shown in an experiment on topic segmentation of broadcast news.","subitem_description_type":"Other"}]},"item_4_biblio_info_10":{"attribute_name":"書誌情報","attribute_value_mlt":[{"bibliographicPageEnd":"41","bibliographic_titles":[{"bibliographic_title":"情報処理学会研究報告知能と複雑系（ICS）"}],"bibliographicPageStart":"37","bibliographicIssueDates":{"bibliographicIssueDate":"2005-03-14","bibliographicIssueDateType":"Issued"},"bibliographicIssueNumber":"24(2004-ICS-139)","bibliographicVolumeNumber":"2005"}]},"relation_version_is_last":true,"weko_creator_id":"1"},"id":50237,"updated":"2025-01-22T07:43:02.682481+00:00","links":{},"created":"2025-01-18T23:15:03.673457+00:00"}