非連續語音傳輸系統

Abstract

This work is aimed to benefit from the concept of discontinuous transmission by sending the DM-encoded speech packets only during talkspurts. We first analyze the effect of transmission errors for EDM coding of autoregressive sources, and we substantiate the theoretical values with simulation values. The combined use of bit alterating rate and prediction gain allows the implementation of a voice activity detector that accurately distinguish between active and silent periods. Furthermore, a comfort noise generator is used to produce on the receiver side synthetic background noise which is not transmitted during the silent periods. This task is done by using the random noise source to excite an all-pole synthesis filter which is characterized by both the short-term energy and the linear prediction parameters of background noise signal.