| Title: | Bottleneck ANN: dealing with small amount of data in shift-MLLR adaptation |
| Other Titles: | Bottleneck ANN: shift-MLLR adaptace pro malé množství dat |
| Authors: | Zajíc, Zbyněk Machlica, Lukáš Müller, Luděk |
| Citation: | ZAJÍC, Zbyněk; MACHLICA, Lukáš; MüLLER, Luděk. Bottleneck ANN: dealing with small amount of data in shift-MLLR adaptation. In: IEEE 11th international conference on signal processing. Beijing: IEEE Press, 2012, p. 507-510. |
| Issue Date: | 2012 |
| Publisher: | IEEE Press |
| Document type: | článek article |
| URI: | http://www.kky.zcu.cz/cs/publications/ZbynekZajic_2012_BottleneckANN http://hdl.handle.net/11025/16950 |
| Keywords: | ASR;adaptace;shift-MLLR;ANN;bottleneck |
| Keywords in different language: | ASR;adaptation;shift-MLLR;ANN;bottleneck |
| Abstract: | Článek popisuje řešení nedostatečného množství dat pro shift-MLLR adaptaci pomocí neuonové sítě bottelneck. |
| Abstract in different language: | The aim of this work is to propose a refinement of the shift-MLLR (shift Maximum Likelihood Linear Regression) adaptation of an acoustics model in the case of limited amount of adaptation data, which can lead to ill-conditioned transformations matrices. We try to suppress the influence of badly estimated transformation parameters utilizing the bottleneck Artificial Neural Network (ANN). The ill-conditioned shift-MLLR transformation is propagated through a bottleneck ANN (suitably trained beforehand), and the output of the net is used as the new refined transformation. To train the ANN the well and the badly conditioned shift-MLLR transformations are used as outputs and inputs of ANN, respectively. |
| Rights: | © Zbyněk Zajíc - Lukáš Machlica - Luděk Müller |
| Appears in Collections: | Články / Articles (KKY) Články / Articles (NTIS) |
Files in This Item:
| File | Description | Size | Format | |
|---|---|---|---|---|
| ZbynekZajic_2012_BottleneckANN.pdf | Plný text | 168,33 kB | Adobe PDF | View/Open |
Please use this identifier to cite or link to this item:
http://hdl.handle.net/11025/16950Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.