DiSS-LPSS Joint Workshop 2010

The 5th Workshop on Disfluency in Spontaneous Speech
The 2nd International Symposium on Linguistic Patterns in Spontaneous Speech

Tokyo, Japan, September 25-26, 2010

Voice Activity Detection based on Combination of Weighted Sub-band Features using Auto-Correlation Function

Kun-Ching Wang (1), Chiun-Li Chin (2), Yi-Hsing Tsai (3)

(1) Department of Information Technology & Communication, Shin Chien University
(2) Department of Applied Information Sciences, Chung Shan Medical University
(3) Information & Communications Research Laboratories, Industrial Technology Research Institute

This paper shows the voice activity detection (VAD) based on combination of weighted sub-band features using autocorrelation function. According to the fact that the noise corruption on each sub-band is different from each other, so the estimated signal to noise ratio (SNR) is employed to weight utility rate of each frequency sub-band. Furthermore, a strategy of sub-band features combination is used to integrate all of weighted sub-band auto-correlation function feature parameter and to develop the combined feature parameter. Experimental results demonstrate that the proposed VAD achieves better performance than existing standard VADs at any noise level.

Index Terms. voice activity detection, auto-correlation, wavelet packet transform, sub-band weighting, feature combination

Full Paper

Bibliographic reference.  Wang, Kun-Ching / Chin, Chiun-Li / Tsai, Yi-Hsing (2010): "Voice activity detection based on combination of weighted sub-band features using auto-correlation function", In DiSS-LPSS-2010, 85-88.