Simulating Native Speaker Shadowing for Non-Native Speech Assessment with Latent Speech Representation

• Item type: SIG Technical Reports(1)
• 公開日: 2024-12-05
• タイトル: Simulating Native Speaker Shadowing for Non-Native Speech Assessment with Latent Speech Representation
• タイトル / 言語: en
• タイトル: Simulating Native Speaker Shadowing for Non-Native Speech Assessment with Latent Speech Representation
• 言語: eng
• キーワード / 主題Scheme: Other
• キーワード / 主題: 話者認識・音声分析
• 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: Haopeng, Geng; Daisuke, Saito; Nobuaki, Minematsu
• 著者名(英): Haopeng, Geng; Daisuke, Saito; Nobuaki, Minematsu
• 論文抄録 / 内容記述タイプ: Other
• 論文抄録 / 内容記述: Evaluating non-native speakers' speech intelligibility is essential in computer-assisted language learning (CALL). Conventional methods, such as using word error rates (WER) from automatic speech recognition (ASR), often lack pedagogical validity due to differences from human speech recognition (HSR). A more effective approach involves native (L1) speakers shadowing non-native (L2) speech, where breakdowns in L1 shadowing indicate unintelligible segments in L2 speech. In this study, we propose a speech generation system that simulates L1 shadowing of L2 speech using seq2seq voice conversion (VC) with latent speech representations. Our experimental results demonstrate that this method effectively replicates the L1 shadowing process, offering a novel tool for evaluating L2 speech intelligibility.
• 論文抄録(英) / 内容記述タイプ: Other
• 論文抄録(英) / 内容記述: Evaluating non-native speakers' speech intelligibility is essential in computer-assisted language learning (CALL). Conventional methods, such as using word error rates (WER) from automatic speech recognition (ASR), often lack pedagogical validity due to differences from human speech recognition (HSR). A more effective approach involves native (L1) speakers shadowing non-native (L2) speech, where breakdowns in L1 shadowing indicate unintelligible segments in L2 speech. In this study, we propose a speech generation system that simulates L1 shadowing of L2 speech using seq2seq voice conversion (VC) with latent speech representations. Our experimental results demonstrate that this method effectively replicates the L1 shadowing process, offering a novel tool for evaluating L2 speech intelligibility.
• 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-154, 号 42, p. 1-6, 発行日 2024-12-05
• 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