| Presentation | 2021-03-04 Optimization source-filtere based speech waveform generation using adversarial training Hayato Mitsui, Yosuke Sugiura, Nozomiko Yasui, Tetsuya Shimamura, |
|---|---|
| PDF Download Page |  PDF download Page Link |
| Abstract(in Japanese) | (See Japanese page) |
| Abstract(in English) | This research aims to improve the accuracy of the source-filter based speech waveform generation model using deep learning. While the source-filter based speech waveform generation model can be implemented with lower computational cost compared with WaveNet based on Pixel CNN, this model produces a low-quality speech. To maintain the naturalness of the generated speech, we introduce a mutli-task training architecture using the adversarial training. In the proposed method, we use the architecture of MelGAN as the adversarial training. From the experimental results, we reveal that the proposed method can obtain the dynamics of speech which was lost in the case of the conventional method. |
| Keyword(in Japanese) | (See Japanese page) |
| Keyword(in English) | Deep Learning / Speech synthesis / Source-Filter theory / Adversarial training |
| Paper # | SIS2020-35 |
| Date of Issue | 2021-02-25 (SIS) |
| Conference Information | |
| Committee | SIS |
|---|---|
| Conference Date | 2021/3/4(2days) |
| Place (in Japanese) | (See Japanese page) |
| Place (in English) | Online |
| Topics (in Japanese) | (See Japanese page) |
| Topics (in English) | Soft Computing, etc. |
| Chair | Noriaki Suetake(Yamaguchi Univ.) |
| Vice Chair | Tomoaki Kimura(Kanagawa Inst. of Tech.) / Naoto Sasaoka(Tottori Univ.) |
| Secretary | Tomoaki Kimura(Kindai Univ.) / Naoto Sasaoka(National Inst. of Tech., Ube College) |
| Assistant | Yukihiro Bandoh(NTT) / Soh Yoshida(Kansai Univ.) |
| Paper Information | |
| Registration To | Technical Committee on Smart Info-Media Systems |
|---|---|
| Language | JPN |
| Title (in Japanese) | (See Japanese page) |
| Sub Title (in Japanese) | (See Japanese page) |
| Title (in English) | Optimization source-filtere based speech waveform generation using adversarial training |
| Sub Title (in English) | |
| Keyword(1) | Deep Learning |
| Keyword(2) | Speech synthesis |
| Keyword(3) | Source-Filter theory |
| Keyword(4) | Adversarial training |
| 1st Author's Name | Hayato Mitsui |
| 1st Author's Affiliation | Graduate School of Science and Engineering, Saitama University(Saitama Univ.) |
| 2nd Author's Name | Yosuke Sugiura |
| 2nd Author's Affiliation | Graduate School of Science and Engineering, Saitama University(Saitama Univ.) |
| 3rd Author's Name | Nozomiko Yasui |
| 3rd Author's Affiliation | Graduate School of Science and Engineering, Saitama University(Saitama Univ.) |
| 4th Author's Name | Tetsuya Shimamura |
| 4th Author's Affiliation | Graduate School of Science and Engineering, Saitama University(Saitama Univ.) |
| Date | 2021-03-04 |
| Paper # | SIS2020-35 |
| Volume (vol) | vol.120 |
| Number (no) | SIS-415 |
| Page | pp.pp.1-4(SIS), |
| #Pages | 4 |
| Date of Issue | 2021-02-25 (SIS) |