| Presentation | 2016-07-28 On the Use of Speaker Codes for Multi-Speaker Modeling in DNN-based Speech Synthesis Nobukatsu Hojo, Yusuke Ijima, Hideyuki Mizuno, |
|---|---|
| PDF Download Page |  PDF download Page Link |
| Abstract(in Japanese) | (See Japanese page) |
| Abstract(in English) | Recent studies have shown that DNN-based speech synthesis can generate more natural synthesized speech than the conventional HMM-based speech synthesis.However, it is not revealed whether the synthesized speech quality can be improved by utilizing a multi-speaker speech corpus.To investigate this problem, this paper proposes DNN-based speech synthesis using speaker codes as a simple method to improve the performance of the conventional speaker dependent DNN-based method. In order to model speaker variation in the DNN, speaker codes are fed to the hidden layer(s) of the DNN. The proposed method trains connection weights of the whole DNN utilizing a multi-speaker speech corpus. To synthesize a speech parameter sequence, a target speaker is chosen from the corpus and the speaker code corresponding to the selected target speaker is fed to the DNN to generate the speaker's voice. We investigated the relationship between the prediction performance and architecture of the DNNs by changing the hidden layer used for speaker codes. Experimental results showed that the proposed model outperformed the conventional speaker-dependent DNN when the model architecture was set at optimal depending on the amount of training data of the selected target speaker. |
| Keyword(in Japanese) | (See Japanese page) |
| Keyword(in English) | speech synthesis / acoustic model / deep neural network / speaker codes |
| Paper # | SP2016-22 |
| Date of Issue | 2016-07-21 (SP) |
| Conference Information | |
| Committee | SP / IPSJ-SLP |
|---|---|
| Conference Date | 2016/7/28(3days) |
| Place (in Japanese) | (See Japanese page) |
| Place (in English) | Takinoyu Hotel |
| Topics (in Japanese) | (See Japanese page) |
| Topics (in English) | Speech recognition and understanding, dialog system, etc. |
| Chair | Kazunori Mano(Shibaura Inst. of Tech.) / Nobuaki Minematsu(Univ. of Tokyo) |
| Vice Chair | Hiroki Mori(Utsunomiya Univ.) |
| Secretary | Hiroki Mori(Kobe Univ.) / (Shizuoka Univ.) |
| Assistant | Taichi Asami(NTT) / Kei Hashimoto(Nagoya Inst. of Tech.) |
| Paper Information | |
| Registration To | Technical Committee on Speech / Special Interest Group on Spoken Language Processing |
|---|---|
| Language | JPN |
| Title (in Japanese) | (See Japanese page) |
| Sub Title (in Japanese) | (See Japanese page) |
| Title (in English) | On the Use of Speaker Codes for Multi-Speaker Modeling in DNN-based Speech Synthesis |
| Sub Title (in English) | |
| Keyword(1) | speech synthesis |
| Keyword(2) | acoustic model |
| Keyword(3) | deep neural network |
| Keyword(4) | speaker codes |
| 1st Author's Name | Nobukatsu Hojo |
| 1st Author's Affiliation | Nippon Telegraph and Telephone Corporation(NTT) |
| 2nd Author's Name | Yusuke Ijima |
| 2nd Author's Affiliation | Nippon Telegraph and Telephone Corporation(NTT) |
| 3rd Author's Name | Hideyuki Mizuno |
| 3rd Author's Affiliation | Tokyo University of Science, Suwa(Tokyo University of Science, Suwa) |
| Date | 2016-07-28 |
| Paper # | SP2016-22 |
| Volume (vol) | vol.116 |
| Number (no) | SP-165 |
| Page | pp.pp.13-18(SP), |
| #Pages | 6 |
| Date of Issue | 2016-07-21 (SP) |