In Internet audio applications, delay and delay jitter affect applications' quality of service mostly. Since packet delays are different and changing over time, the receiver needs to buffer some amount of packets before playout. Therefore, the amount of buffered packets and timing of playout are very important for the performance of the applications. Here we adopt an auto-regressive (AR) model for estimation of packet delay and deploy a robust identification algorithm for adjustment of parameters of AR process. In our preliminary experiments, this robust algorithm leads better performance when the noise is correlated and/or non-stationary, and also it is robust to model uncertainties.