Acoustic Modeling Using Restricted Boltzmann Machines and Deep Belief Networks for Statistical Parametric Speech Synthesis and Voice Conversion

• Item type: SIG Technical Reports(1)
• 公開日: 2013-12-12
• タイトル: Acoustic Modeling Using Restricted Boltzmann Machines and Deep Belief Networks for Statistical Parametric Speech Synthesis and Voice Conversion
• タイトル / 言語: en
• タイトル: Acoustic Modeling Using Restricted Boltzmann Machines and Deep Belief Networks for Statistical Parametric Speech Synthesis and Voice Conversion
• 言語: eng
• キーワード / 主題Scheme: Other
• キーワード / 主題: 招待講演
• 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: National Engineering Laboratory for Speech and Language Information Processing University of Science and Technology of China
• 著者所属: National Engineering Laboratory for Speech and Language Information Processing University of Science and Technology of China
• 著者所属: National Engineering Laboratory for Speech and Language Information Processing University of Science and Technology of China
• 著者所属(英): en
• 著者所属(英): National Engineering Laboratory for Speech and Language Information Processing University of Science and Technology of China
• 著者所属(英): en
• 著者所属(英): National Engineering Laboratory for Speech and Language Information Processing University of Science and Technology of China
• 著者所属(英): en
• 著者所属(英): National Engineering Laboratory for Speech and Language Information Processing University of Science and Technology of China
• 著者名: Zhen-HuaLing; Ling-HuiChen; Li-RongDai
• 著者名(英): Zhen-Hua, Ling; Ling-Hui, Chen; Li-Rong, Dai
• 論文抄録 / 内容記述タイプ: Other
• 論文抄録 / 内容記述: This paper summarizes our previous work on spectral modeling using restricted Boltzmann machines (RBM) and deep belief networks (DBN) for statistical parametric speech synthesis and voice conversion. This approach improves the conventional methods in two ways. First, the raw spectral envelopes extracted by the STRAIGHT vocoder are used as the features for spectral modeling. Second, instead of using single Gaussian distribution, we adopt RBMs or DBNs to represent the distribution of the envelopes at each HMM state or GMM mixture. Our experimental results show the effectiveness of this proposed method in improving the naturalness and similarity of the generated speech.
• 論文抄録(英) / 内容記述タイプ: Other
• 論文抄録(英) / 内容記述: This paper summarizes our previous work on spectral modeling using restricted Boltzmann machines (RBM) and deep belief networks (DBN) for statistical parametric speech synthesis and voice conversion. This approach improves the conventional methods in two ways. First, the raw spectral envelopes extracted by the STRAIGHT vocoder are used as the features for spectral modeling. Second, instead of using single Gaussian distribution, we adopt RBMs or DBNs to represent the distribution of the envelopes at each HMM state or GMM mixture. Our experimental results show the effectiveness of this proposed method in improving the naturalness and similarity of the generated speech.
• 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 2013-SLP-99, 号 17, p. 1-6, 発行日 2013-12-12
• 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