In　order　to　improve　the　robustness　of　voice　activity　detection　(VAD),　the　use　of　an　algorithm　based　on　complex　Gaussian　mixture　model　under　nonstationary　noisy　environments　was　presented.　In　the　algorithm,　the　clean　speech　distribution　was　modelled　by　complex　Gaussian　mixture　model,　and　the　a　priori　SNR　was　estimated　based　on　the　pre-trained　complex　Gaussian　mixture　model.　The　introduction　of　complex　Gaussian　mixture　model　not　only　improved　the　performance　of　voice　activity　detection,　but　also　avoided　the　estimation　of　a　priori　SNR　using　minimum　mean　square　error　short　spectral　amplitude　estimator.　The　system　performance　under　noisy　environments　was　evaluated　using　NOISEX-92　database.　Experimental　results　show　that　the　algorithm　can　work　more　robustly　under　nonstationary　noisy　environments.