Comparison between DFT- and DWT-based speech/non-speech detection for adverse environments

Abstract

The goal of this paper is to evaluate the wavelet/frequency-based voice activity detection (VAD) algorithms under harsh conditions. A new frequency-based speech classifier has been developed based on a single subband distance feature in cooperating with adaptive percentile filter. Experimental results in clean, noisy and reverberant environments are provided. Results show that: (i) the group of algorithms exploiting the subband power distance feature mostly outperforms the state-of-the-art VAD standardized for the G. 729 B, the ETSI AFE ES 202 050 in terms of classification measures; (ii) the robustness of the model-based VAD methods still holds in a completely mismatched reverberant environment.