Effective speech communication can be achieved by taking the speech signal when microphone is active and suppressing the noise when it is passive. The model we proposed in this paper is to take Feature vectors of pre-defined speech and noise signal’s, which already are stored for processing. Then centriods of the Feature Vectors were calculated using k-means algorithm. The Minimum distance between framed feature vectors of input signal and centriods of pre-defined signals are estimated using Euclidian distance. The ratio obtained between noise and speech minimum distance vectors will represent the voice activity at the microphone. The evaluation of the ratio indicates the significant performance of voice activity detection in noisy environment as well.