Enhancing Feature Integration to Improve Classification Accuracy of Similar Categories in Acoustic Scene Classification

• Item type: SIG Technical Reports(1)
• 公開日: 2024-06-07
• タイトル: Enhancing Feature Integration to Improve Classification Accuracy of Similar Categories in Acoustic Scene Classification
• タイトル / 言語: en
• タイトル: Enhancing Feature Integration to Improve Classification Accuracy of Similar Categories in Acoustic Scene Classification
• 言語: eng
• キーワード / 主題Scheme: Other
• キーワード / 主題: ポスターセッション2
• 資源タイプ / 資源タイプ識別子: http://purl.org/coar/resource_type/c_18gh
• 資源タイプ: technical report
• 著者所属: The University of Tokyo
• 著者所属: The University of Tokyo
• 著者所属: The University of Tokyo
• 著者所属(英): en
• 著者所属(英): The University of Tokyo
• 著者所属(英): en
• 著者所属(英): The University of Tokyo
• 著者所属(英): en
• 著者所属(英): The University of Tokyo
• 著者名: Shuting, Hao; Daisuke, Saito; Nobuaki, Minematsu
• 著者名(英): Shuting, Hao; Daisuke, Saito; Nobuaki, Minematsu
• 論文抄録 / 内容記述タイプ: Other
• 論文抄録 / 内容記述: This study focuses on Acoustic Scene Classification (ASC), which categorizes environmental audio streams into predefined semantic labels. We introduce a novel architecture that integrates multi-layer classifiers and direct finetuning, presenting a new perspective in ASC research. The study employs the TAU Urban Acoustic Scenes 2022 Mobile dataset for fine-tuning and validation. We utilized the SSAST model, pre-trained on the AudioSet and LibriSpeech datasets, and fine-tuned it on the TAU dataset with a unique approach to enhance ASC-specific feature learning. Our layered SSAST system achieved an accuracy of 52.17% and an AUC of 88.66% in ASC, marking a notable improvement over the baseline with absolute increases of 0.99% in accuracy and 0.85% in AUC.
• 論文抄録(英) / 内容記述タイプ: Other
• 論文抄録(英) / 内容記述: This study focuses on Acoustic Scene Classification (ASC), which categorizes environmental audio streams into predefined semantic labels. We introduce a novel architecture that integrates multi-layer classifiers and direct finetuning, presenting a new perspective in ASC research. The study employs the TAU Urban Acoustic Scenes 2022 Mobile dataset for fine-tuning and validation. We utilized the SSAST model, pre-trained on the AudioSet and LibriSpeech datasets, and fine-tuned it on the TAU dataset with a unique approach to enhance ASC-specific feature learning. Our layered SSAST system achieved an accuracy of 52.17% and an AUC of 88.66% in ASC, marking a notable improvement over the baseline with absolute increases of 0.99% in accuracy and 0.85% in AUC.
• 書誌レコードID / 収録物識別子タイプ: NCID
• 書誌レコードID / 収録物識別子: AN10442647
• 書誌情報: 研究報告音声言語情報処理（SLP）; 巻 2024-SLP-152, 号 53, p. 1-5, 発行日 2024-06-07
• ISSN / 収録物識別子タイプ: ISSN
• ISSN / 収録物識別子: 2188-8663
• Notice: SIG Technical Reports are nonrefereed and hence may later appear in any journals, conferences, symposia, etc.
• 出版者 / 言語: ja
• 出版者: 情報処理学会