Presentation | 2017-03-01 [Poster Presentation] An investigation of speaker adaptation method for DNN-based speech synthesis using speaker codes Nobukatsu Hojo, Yusuke Ijima, |
---|---|
PDF Download Page | PDF download Page Link |
Abstract(in Japanese) | (See Japanese page) |
Abstract(in English) | In this work, we conducted objective evaluation experiments on the conventional speaker adaptation methods for DNN-based text-to-speech synthesis in order to compare their performance. One speaker adaptation method is based on speaker codes, which uses speaker codes as a input to the hidden layers. Besides the speaker adaptation method which reestimates the target speaker’s code, we have proposed the adaptation method to estimate the regression function for speaker codes. The other speaker adaptation method uses speaker dependent output layer, which shares hidden layers across all the speakers while the output layers are composed of speaker-dependent nodes. The model can be adapted to a target speaker by estimating regression layer parameters for the corresponding output layer. It has not been revealed how the difference of the model architecture and the number of adaptation parameters affects the performance. This work conducts speaker adaptation experiments using speech corpus which consists of 35 speakers’ speech data, and compared the performance by objective measure. From the experimental results, it was revealed that the speaker code-based models with more adaptation parameters show higher performance. It was also shown that the speaker code based models show comparable performance to speaker dependent output layer models when the number of adaptation parameters is set to large. |
Keyword(in Japanese) | (See Japanese page) |
Keyword(in English) | speech synthesis / acoustic model / deep neural network / speaker codes |
Paper # | EA2016-108,SIP2016-163,SP2016-103 |
Date of Issue | 2017-02-22 (EA, SIP, SP) |
Conference Information | |
Committee | SP / SIP / EA |
---|---|
Conference Date | 2017/3/1(2days) |
Place (in Japanese) | (See Japanese page) |
Place (in English) | Okinawa Industry Support Center |
Topics (in Japanese) | (See Japanese page) |
Topics (in English) | Speech, Engineering/Electro Acoustics, Signal Processing, and Related Topics |
Chair | Kazunori Mano(Shibaura Inst. of Tech.) / Makoto Nakashizuka(Chiba Inst. of Tech.) / Mitsunori Mizumachi(Kyushu Inst. of Tech.) |
Vice Chair | Hiroki Mori(Utsunomiya Univ.) / Masahiro Okuda(Univ. of Kitakyushu) / Shogo Muramatsu(Niigata Univ.) / Yoichi Haneda(Univ. of Electro-Comm.) / Suehiro Shimauchi(NTT) |
Secretary | Hiroki Mori(Kobe Univ.) / Masahiro Okuda(Shizuoka Univ.) / Shogo Muramatsu(Ritsumeikan Univ.) / Yoichi Haneda(Chiba Inst. of Tech.) / Suehiro Shimauchi(KDDI R&D Labs.) |
Assistant | Taichi Asami(NTT) / Kei Hashimoto(Nagoya Inst. of Tech.) / Osamu Watanabe(Takushoku Univ.) / Shigeto Takeoka(Shizuoka Inst. of Science and Tech.) / TREVINO Jorge(Tohoku Univ.) |
Paper Information | |
Registration To | Technical Committee on Speech / Technical Committee on Signal Processing / Technical Committee on Engineering Acoustics |
---|---|
Language | JPN |
Title (in Japanese) | (See Japanese page) |
Sub Title (in Japanese) | (See Japanese page) |
Title (in English) | [Poster Presentation] An investigation of speaker adaptation method for DNN-based speech synthesis using speaker codes |
Sub Title (in English) | |
Keyword(1) | speech synthesis |
Keyword(2) | acoustic model |
Keyword(3) | deep neural network |
Keyword(4) | speaker codes |
Keyword(5) | |
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) |
Date | 2017-03-01 |
Paper # | EA2016-108,SIP2016-163,SP2016-103 |
Volume (vol) | vol.116 |
Number (no) | EA-475,SIP-476,SP-477 |
Page | pp.pp.147-152(EA), pp.147-152(SIP), pp.147-152(SP), |
#Pages | 6 |
Date of Issue | 2017-02-22 (EA, SIP, SP) |