Voice activity detection based on minimum statistics

In this study, a voice activity detection method based on minimum statistics is developed. The method is suitable for use in non-stationary noise environments, and it also has low computational load. Noise energy is estimated by tracking the minimum of the noisy signal energy on a finite window and the decision rule is constructed accordingly. A number of experiments is applied to the system which is combined with an adaptive threshold structure and it is observed that it provides better results than the voice activity detection standards especially in non-stationary noise.