Speaking Rate Control by HiFi-GAN using Feature Interpolation

• Item type: SIG Technical Reports(1)
• 公開日: 2022-02-22
• タイトル: Speaking Rate Control by HiFi-GAN using Feature Interpolation
• タイトル / 言語: en
• タイトル: Speaking Rate Control by HiFi-GAN using Feature Interpolation
• 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: National Institute of Information and Communications Technology
• 著者所属: National Institute of Information and Communications Technology
• 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): National Institute of Information and Communications Technology
• 著者所属(英): en
• 著者所属(英): National Institute of Information and Communications Technology
• 著者所属(英): en
• 著者所属(英): The University of Tokyo
• 著者名: Detai, Xin; Shinnosuke, Takamichi; Takuma, Okamoto; Hisashi, Kawai; Hiroshi, Saruwatari
• 著者名(英): Detai, Xin; Shinnosuke, Takamichi; Takuma, Okamoto; Hisashi, Kawai; Hiroshi, Saruwatari
• 論文抄録 / 内容記述タイプ: Other
• 論文抄録 / 内容記述: We investigate the possibility of controlling speaking rate by using the HiFi-GAN neural vocoder. Although traditional time-scale modification (TSM) algorithms have been widely applied in real-world applications, their performance and efficiency are relatively low. Recent work of neural vocoder has shown the possibility of synthesizing speech with high fidelity and efficiency. The proposed method inserts an interpolation layer into the HiFi-GAN to control the speaking rate. A signal resampling method and an image scaling method are implemented in the proposed method to warp the mel-spectrogram or hidden features of the neural vocoder. We also design a Japanese speech corpus to evaluate the proposed speaking rate control method. Experimental results of comprehensive objective and subjective evaluations demonstrate that the proposed method can control speaking rate with higher quality and efficiency than a baseline TSM algorithm. We open-source the corpus and give future directions of speaking rate control by a neural vocoder.
• 論文抄録(英) / 内容記述タイプ: Other
• 論文抄録(英) / 内容記述: We investigate the possibility of controlling speaking rate by using the HiFi-GAN neural vocoder. Although traditional time-scale modification (TSM) algorithms have been widely applied in real-world applications, their performance and efficiency are relatively low. Recent work of neural vocoder has shown the possibility of synthesizing speech with high fidelity and efficiency. The proposed method inserts an interpolation layer into the HiFi-GAN to control the speaking rate. A signal resampling method and an image scaling method are implemented in the proposed method to warp the mel-spectrogram or hidden features of the neural vocoder. We also design a Japanese speech corpus to evaluate the proposed speaking rate control method. Experimental results of comprehensive objective and subjective evaluations demonstrate that the proposed method can control speaking rate with higher quality and efficiency than a baseline TSM algorithm. We open-source the corpus and give future directions of speaking rate control by a neural vocoder.
• 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 2022-SLP-140, 号 32, p. 1-6, 発行日 2022-02-22
• 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