Voice activity detection based on using wavelet packet |
| |
Authors: | Mohadese Eshaghi MR Karami Mollaei |
| |
Affiliation: | University of Technology, Faculty of Electrical and Computer Engineering, P.O. Box 47135-484, Babol, Iran |
| |
Abstract: | This paper, presents a robust voice activity detection (VAD) technique based on wavelet packet. In this technique sub-bands and their amplitudes are represented as the vectors for each sample time in order to find a new feature from the frequency and amplitude changes. On the other hand, the multi-resolution analysis property of the wavelet packet transform (WPT), the voiced, unvoiced, and transient components of speech can be distinctly discriminated. Then, a new feature extraction method is implemented based on observations of the angles between vectors. This feature extraction method retains most unvoiced sounds in a voice active frame. Experimental results show that the proposed WT feature parameter can extract the speech activity under poor SNR conditions and that it is also insensitive to variable-level of noise. |
| |
Keywords: | |
本文献已被 ScienceDirect 等数据库收录! |
|