Cross-lingual Singing Voice Conversion Leveraging Transformer-based ASR Encoder

• Item type: SIG Technical Reports(1)
• 公開日: 2024-06-07
• タイトル: Cross-lingual Singing Voice Conversion Leveraging Transformer-based ASR Encoder
• タイトル / 言語: en
• タイトル: Cross-lingual Singing Voice Conversion Leveraging Transformer-based ASR Encoder
• 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: Graduate School of Engineering, The University of Tokyo
• 著者所属: Graduate School of Engineering, The University of Tokyo
• 著者所属: Graduate School of Engineering, The University of Tokyo
• 著者所属(英): en
• 著者所属(英): Graduate School of Engineering, The University of Tokyo
• 著者所属(英): en
• 著者所属(英): Graduate School of Engineering, The University of Tokyo
• 著者所属(英): en
• 著者所属(英): Graduate School of Engineering, The University of Tokyo
• 著者名: Xinyuan, Zhao; Daisuke, Saito; Nobuaki, Minematsu
• 著者名(英): Xinyuan, Zhao; Daisuke, Saito; Nobuaki, Minematsu
• 論文抄録 / 内容記述タイプ: Other
• 論文抄録 / 内容記述: In this paper, we focus on singing voice conversion task, including monolingual and cross-lingual singing voice conversion. We proposed a method to modify MaskCycleGAN, a model designed for nonparallel speech conversion task, and applied the proposed model for the singing voice conversion task. The new model concatenates MaskCycleGAN with Whisper, a transformer-based ASR encoder that recognizes the content of the song. The experimental results demonstrated significant performance improvements over the original MaskCycleGAN in both monolingual and cross-lingual scenarios for the singing voice conversion task.
• 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 47, p. 1-4, 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