| Presentation | 2014-12-16 Deep neural network-based feature transformation for reverberant speaker identification Zhaofeng Zhang, Longbiao Wang, Atsuhiko Kai, Weifeng Li, Masahiro Iwahashi, |
|---|---|
| PDF Download Page |  PDF download Page Link |
| Abstract(in Japanese) | (See Japanese page) |
| Abstract(in English) | Deep neural network has been shown to be effective in many automatic speech recognition. In this paper, we proposed both bottleneck-DNN (BF-DNN) and denoising-autoencoder (DAE) based feature transformation for reverberant speaker identification. For the BF-DNN, we consider that the DNNs can transform the speech frame to a new space with gerater discriminative classification ability for speaker identification. While the DAE dereverberation suppresses the reverberation can improve the performance of speaker identification. Since the BF-DNN and DAE have a great complementary nature, the linear likelihood combination of these two methods is expected be effective for this task. The evaluation experiment shows proposed method performed better than conventional method in reverberant environment. Our proposed method outperforms the multichannel least mean squares. The relative error reduction rate are 21.4% for BF-DNN and 47.0% for DAE, respectively. Moreover, the combination of this two methods further improved the performance. |
| Keyword(in Japanese) | (See Japanese page) |
| Keyword(in English) | Deep neural network / Bottleneck feature / Denoising autoencoder / Speaker identification / Dereverberation / Feature transformation |
| Paper # | SP2014-119 |
| Date of Issue |
| Conference Information | |
| Committee | SP |
|---|---|
| Conference Date | 2014/12/8(1days) |
| Place (in Japanese) | (See Japanese page) |
| Place (in English) | |
| Topics (in Japanese) | (See Japanese page) |
| Topics (in English) | |
| Chair | |
| Vice Chair | |
| Secretary | |
| Assistant | |
| Paper Information | |
| Registration To | Speech (SP) |
|---|---|
| Language | JPN |
| Title (in Japanese) | (See Japanese page) |
| Sub Title (in Japanese) | (See Japanese page) |
| Title (in English) | Deep neural network-based feature transformation for reverberant speaker identification |
| Sub Title (in English) | |
| Keyword(1) | Deep neural network |
| Keyword(2) | Bottleneck feature |
| Keyword(3) | Denoising autoencoder |
| Keyword(4) | Speaker identification |
| Keyword(5) | Dereverberation |
| Keyword(6) | Feature transformation |
| 1st Author's Name | Zhaofeng Zhang |
| 1st Author's Affiliation | Nagaoka University of Technology() |
| 2nd Author's Name | Longbiao Wang |
| 2nd Author's Affiliation | Nagaoka University of Technology |
| 3rd Author's Name | Atsuhiko Kai |
| 3rd Author's Affiliation | Shizuoka University |
| 4th Author's Name | Weifeng Li |
| 4th Author's Affiliation | Tsinghua University |
| 5th Author's Name | Masahiro Iwahashi |
| 5th Author's Affiliation | Nagaoka University of Technology |
| Date | 2014-12-16 |
| Paper # | SP2014-119 |
| Volume (vol) | vol.114 |
| Number (no) | 365 |
| Page | pp.pp.- |
| #Pages | 6 |
| Date of Issue | |