We describe the design, implementation and performance of a stand-alone Media Server. This server is intended to be used in Voice Over IP Systems for audio interaction with users, NAT traversal, call monitoring and media recording. Its advantage is that it is application independent, which means that it can be used by any service and for any signaling protocol.

In this paper, we describe the design of the server, of its control protocol and of its functionalities such as file playing, DTMF detection, RTP proxying and audio recording. For file playing, we propose an approach to schedule RTP packet transmission that overcomes the irregularity of event scheduling of the operating system and for conversation recording, we propose an original approach to synchronize media streams that allows to avoid voice overlap while recording a conversation.

We have conducted many experiments on the Media Server. We evaluated the accuracy of the file playing timing under different load conditions, the effect of the RTP proxying on the media quality and the subjective quality of recorded conversations. The results of these experiments show the great impact of the load of the Media Server on the performance of its functionalities, the improvement that the packet scheduling algorithm brings to the precision of inter-departure timing, and the good subjective quality of the recorded conversation using our synchronizing approach.