A low complexity statistical voice activity detector with performance comparisons to ITU-T/ETSI voice activity detectors

Traditionally voice activity detection algorithms are based on any combination of general speech properties such as temporal energy variations, periodicity, and spectrum. This paper describes a statistical method for voice activity detection using a signal-to-noise ratio measure. The method employs a low-variance spectrum estimate and an adaptive threshold to make a statistical voice activity decision. An added advantage of the method is that it only requires low resources such as memory and computational time. Furthermore the method has been compared to modern standard voice activity detection algorithms with results indicating good performance in babble, white and vehicle noise.