The Robustness Study of Multiple Kernel Learning Approaches for VAD

DOI

10.2991/meici-18.2018.150How to use a DOI?

Keywords

Voice activity detection; Deep learning; Multiple Kernel Learning; Robustness

Abstract

Recently, although the MKL-SVM-based VAD has achieved desirable performance, the VAD base on deep learning networks, are attracting greater research interest than their with overwhelming advantages. In this paper, we focus on investigation and analysis the noise robustness of VAD systems multiple-feature-based on MKL-SVM comparing DBN, LSTM and CNN at frame level under various noisy conditions on TIMIT. Experimental results have shown that the MKL-SVM-based VAD not only is not inferior to deep learning networks VADs, but also has a low detection complexity. Further experiment on the information robustness task demonstrates that the MKL-SVM-based VAD apply the advantages of multiple features effectively.

Copyright

© 2018, the Authors. Published by Atlantis Press.

Open Access

This is an open access article distributed under the CC BY-NC license (http://creativecommons.org/licenses/by-nc/4.0/).

Download article (PDF)

TY  - CONF
AU  - Jie Zhang
AU  - Mantao Wang
AU  - Haitao Tang
AU  - Qiang Huang
AU  - Haibo Pu
AU  - Lixin Luo
AU  - Zhihao Zhou
PY  - 2018/12
DA  - 2018/12
TI  - The Robustness Study of Multiple Kernel Learning Approaches for VAD
BT  - Proceedings of the 2018 8th International Conference on Management, Education and Information (MEICI 2018)
PB  - Atlantis Press
SP  - 757
EP  - 763
SN  - 1951-6851
UR  - https://doi.org/10.2991/meici-18.2018.150
DO  - 10.2991/meici-18.2018.150
ID  - Zhang2018/12
ER  -