Presentation 2020-03-06
Adversarial Training using Self-Attention Architecture for Speech Enhancement Network
Yosuke Sugiura, Shimamura Tetsuya,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) In this paper, we propose a new adversarial training for improving performance of the speech enhancement network. In the proposed method, the self-attention architecture is applied to the generated speech waveform to obtain categorical information without supervised training. The categorical information concatenated to the input of the discriminator relaxes the problem and achieves the accurate adversarial training. Through the simulation experiments, we reveal that the proposed method can construct the network that can accurately restore the speech waveform compared with the conventional method.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) Speech Enhancement / Neural Network / Adversarial Training / Self-Attention
Paper # SIS2019-59
Date of Issue 2020-02-27 (SIS)

Conference Information
Committee SIS
Conference Date 2020/3/5(2days)
Place (in Japanese) (See Japanese page)
Place (in English) Saitama Hall
Topics (in Japanese) (See Japanese page)
Topics (in English) Soft Computing, etc.
Chair Takayuki Nakachi(NTT)
Vice Chair Noriaki Suetake(Yamaguchi Univ.) / Tomoaki Kimura(Kanagawa Inst. of Tech.)
Secretary Noriaki Suetake(Tokyo Metropolitan Univ.) / Tomoaki Kimura(Kindai Univ.)
Assistant Hideaki Misawa(National Inst. of Tech., Ube College) / Yukihiro Bandoh(NTT)

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) Adversarial Training using Self-Attention Architecture for Speech Enhancement Network
Sub Title (in English)
Keyword(1) Speech Enhancement
Keyword(2) Neural Network
Keyword(3) Adversarial Training
Keyword(4) Self-Attention
1st Author's Name Yosuke Sugiura
1st Author's Affiliation Saitama University(Saitama Univ.)
2nd Author's Name Shimamura Tetsuya
2nd Author's Affiliation Saitama University(Saitama Univ.)
Date 2020-03-06
Paper # SIS2019-59
Volume (vol) vol.119
Number (no) SIS-458
Page pp.pp.125-129(SIS),
#Pages 5
Date of Issue 2020-02-27 (SIS)