Presentation | 2016-08-25 Diversity-driven Semi-supervised Ensemble DNN Acoustic Model Training Sheng Li, Xugang Lu, Shinsuke Sakai, Tatsuya Kawahara, |
---|---|
PDF Download Page | PDF download Page Link |
Abstract(in Japanese) | (See Japanese page) |
Abstract(in English) | We focus on effective training DNN (Deep Neural Network) acoustic models for Chinese spoken lectures with only limited labeled speech and abundant unlabeled speech. Unlike selectively using the unlabeled data in most semi-supervised DNN training methods and working only under supervised setting in previous ensemble DNN training methods, we work on more generalized ensemble training method for both labeled and unlabeled data. In our proposed method, a pair of models is trained in parallel with diverse labels generated for unlabeled data. Together with the standard cross entropy, the KL divergence between each individual model over unlabeled data is incorporated during training. Experiments show that our proposed method can effectively utilize unlabeled data and outperforms other well-established semi-supervised methods. |
Keyword(in Japanese) | (See Japanese page) |
Keyword(in English) | Speech recognitionAcoustic modelDNNSemi-superivsed trainingEnsemble training |
Paper # | SP2016-40 |
Date of Issue | 2016-08-17 (SP) |
Conference Information | |
Committee | SP |
---|---|
Conference Date | 2016/8/24(2days) |
Place (in Japanese) | (See Japanese page) |
Place (in English) | ACCMS, Kyoto Univ. |
Topics (in Japanese) | (See Japanese page) |
Topics (in English) | Audio event processing, etc. |
Chair | Kazunori Mano(Shibaura Inst. of Tech.) |
Vice Chair | Hiroki Mori(Utsunomiya Univ.) |
Secretary | Hiroki Mori(Kobe Univ.) |
Assistant | Taichi Asami(NTT) / Kei Hashimoto(Nagoya Inst. of Tech.) |
Paper Information | |
Registration To | Technical Committee on Speech |
---|---|
Language | ENG |
Title (in Japanese) | (See Japanese page) |
Sub Title (in Japanese) | (See Japanese page) |
Title (in English) | Diversity-driven Semi-supervised Ensemble DNN Acoustic Model Training |
Sub Title (in English) | |
Keyword(1) | Speech recognitionAcoustic modelDNNSemi-superivsed trainingEnsemble training |
1st Author's Name | Sheng Li |
1st Author's Affiliation | Kyoto University(Kyoto Univ.) |
2nd Author's Name | Xugang Lu |
2nd Author's Affiliation | National Institute of Information and Communications Technology(NICT) |
3rd Author's Name | Shinsuke Sakai |
3rd Author's Affiliation | Kyoto University(Kyoto Univ.) |
4th Author's Name | Tatsuya Kawahara |
4th Author's Affiliation | Kyoto University(Kyoto Univ.) |
Date | 2016-08-25 |
Paper # | SP2016-40 |
Volume (vol) | vol.116 |
Number (no) | SP-189 |
Page | pp.pp.71-76(SP), |
#Pages | 6 |
Date of Issue | 2016-08-17 (SP) |