New speech processing issues in IP telephony

In the field of IP telephony, there are various problems to be dealt with before the QoS currently guaranteed by PSTN networks can be achieved. In this paper, we discuss the main causes of degradation in the quality of conversation in an IP communication scenario. We then point out some new QoS problems involved in speech processing. More specifically, two important aspects of discontinuous transmission are dealt with: the impact of the voice activity detector (VAD) on source throughput and the need for an efficient system of comfort noise. In addition, in a differentiated services (DiffServ) scenario we demonstrate the importance of using a multirate intrastandard codec, together with valid criteria for the reconstruction of lost frames.