A robust voice activity detector applied for AMR

In a digital mobile communication system, during typical telephone conversation, 35% of the time is the real cost to transmit active speech, the other remaining time is full of inactive speech including silence, pauses and noise. Considering the limited spectral resources and power consumption of a mobile phone, a significant capacity gain may be achieved by exploiting the silence gaps. How to develop the silence efficiently for other aims is an important question. The voice activity detector is a key part used to implement silence compression, which can be used to reduce the power consumption of hand-held equipment in the GSM system. In this paper, a new robust voice activity detector based on higher order statistics and applied to a variable bit-rate coder is discussed. It is based on third order cumulants and combined with an adaptive multi-rate (AMR) codec. The trial indicates that the performance of the AMR VAD is obviously improved