The current Internet provides users with a single class best effort service which does not promise anything in terms of performance guarantees. As a result, Real-time audio over IP networks often suffers from varying packet loss rates, delays and available bandwidth. At this time, Forward Error Correction (FEC) seems to be an efficient way to mitigate the impact packet losses. Nevertheless, the use of FEC has two main drawbacks: it increases the bandwidth requirement of the source and it increases the end-to-end delay (a destination typically has to wait longer to decode the FEC as more FEC information is used). Bolot and al. proposed an adaptive rate/error control which optimizes a subjective measure of quality and incorporates the constraints of rate control . This scheme proved to be efficient but it does take into account the addition delay due to the FEC, namely, it does not try to optimize the overall end-to-end delay.

We believe that it is important to take the end-to-end delay into account in the adaptive control scheme for two main reasons. First, it is recognized that above a certain threshold (around 150ms) the end-to-end delay starts to be annoying, or at least noticeable. Second, the emergence of new differentiated services such as the Alternative Best Effort (ABE) offers applications the trade-off between receiving a lower end-to-end delay or more overall throughput.

In our work, we propose an adaptive 'delay aware' error control. We consider the perceived audio quality as a function of the audio encoding rate received at the destination AND of the end-to-end delay. We develop joint rate/error/delay control algorithms which: (1) optimize this measure of audio quality and (2) are TCP-Friendly. In the context of single class best effort networks (also refered as 'Flat' networks), we show that the introduction of this new dimension of delay can influence the way you choose the redundant information. Then, for alternative best effort (ABE) networks, we investigate whether there is a real benefit, for the audio sources, to trade delay for throughput (and packet losses).