| Presentation # | 1 |
| Session: | ASR I |
| Session Time: | Wednesday, December 19, 10:00 - 12:00 |
| Presentation Time: | Wednesday, December 19, 10:00 - 12:00 |
| Presentation: |
Poster
|
| Topic: |
Speech recognition and synthesis: |
| Paper Title: |
HIGH-DEGREE FEATURE FOR DEEP NEURAL NETWORK BASED ACOUSTIC MODEL |
| Authors: |
Hoon Chung; Electronics and Telecommunications Research Institute | | |
| | Sung Joo Lee; Electronics and Telecommunications Research Institute | | |
| | Jeon Gue Park; Electronics and Telecommunications Research Institute | | |
| Abstract: |
In this paper, we propose to use high-degree features using polynomial expansion to improve the discrimination performance of Deep Neural Network (DNN) based acoustic model. Thanks to the success of DNNs for high-dimensional non-linear classification problems, various acoustic information can be represented in high dimensional features, and the non-linear characteristics of speech signal can be robustly generalized in DNN-based acoustic models. Even though it is not clear how DNNs to solve the classification problem, the use of high-dimensional features is based on a well-known knowledge that it helps separability of patters. There is another well-known knowledge that high-degree features increase linear separability of non-linear input features. However, there is little work to exploit high-degree features. Therefore, in this work, we investigate the high-degree features to improve the performance of DNN-based acoustic model further. In this work, the proposed approach was evaluated on a Wall Street Journal (WSJ) speech recognition domain. The proposed method achieved up to 21.8% error reduction rate for the Eval92 test set by reducing the word error rate from 4.82% to 3.77% when using degree-2 polynomial expansion. |