Deep Learning-Based Voice Conversion

• Item type: SIG Technical Reports(1)
• 公開日: 2019-02-20
• タイトル: Deep Learning-Based Voice Conversion
• タイトル / 言語: en
• タイトル: Deep Learning-Based Voice Conversion
• 言語: eng
• 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: University of Science and Technology of China
• 著者所属(英): en
• 著者所属(英): University of Science and Technology of China
• 著者名: Zhenhua, Ling
• 著者名(英): Zhenhua, Ling
• 論文抄録 / 内容記述タイプ: Other
• 論文抄録 / 内容記述: I will introduce our recent work on applying deep learning techniques to voice conversion in this talk. Several methods have been proposed to improve different components in the pipeline of a statistical parametric voice conversion system, including deep neural networks with layer-wise generative training for acoustic modeling, deep autoencoders with binary distributed hidden units for feature representation, and WaveNet vocoder with limited training data for waveform reconstruction. Then, I will introduce our system designed for Voice Conversion Challenge 2018, which achieved the best performance under both parallel and non-parallel conditions in this evaluation. After this, I will present our recent progress on sequence-to-sequence acoustic modeling for voice conversion, which converts the acoustic features and durations of source utterances simultaneously using a unified acoustic model. Finally, some discussions on the future development of voice conversion techniques will be given.
• 論文抄録(英) / 内容記述タイプ: Other
• 論文抄録(英) / 内容記述: I will introduce our recent work on applying deep learning techniques to voice conversion in this talk. Several methods have been proposed to improve different components in the pipeline of a statistical parametric voice conversion system, including deep neural networks with layer-wise generative training for acoustic modeling, deep autoencoders with binary distributed hidden units for feature representation, and WaveNet vocoder with limited training data for waveform reconstruction. Then, I will introduce our system designed for Voice Conversion Challenge 2018, which achieved the best performance under both parallel and non-parallel conditions in this evaluation. After this, I will present our recent progress on sequence-to-sequence acoustic modeling for voice conversion, which converts the acoustic features and durations of source utterances simultaneously using a unified acoustic model. Finally, some discussions on the future development of voice conversion techniques will be given.
• 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 2019-SLP-126, 号 4, p. 1-1, 発行日 2019-02-20
• 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