In Voice over IP applications, the receiver will eventually experience packet losses, either caused by transmission errors (e.g. on wireless links), or when packets are delayed beyond their scheduled playout time. The task of the receiver's packet loss concealment (PLC) algorithm is to generate an estimation for the missing speech frames in order to limit the audible distortion. The achievable quality of a PLC approach depends on different factors, in particular the length of the loss, the stationarity of the signal at that time, and the amount of available information from preceding and following segments (depending on tolerable delay). The task of packet loss concealment is complicated in case highly compressive speech codecs are used which lead to error propagation into following frames.
Our research of optimal PLC algorithms includes signal dependent extra-/interpolation techniques which utilize statistical dependencies to previous and (if available) following frames. The receiver based concealment can be assisted by transmission of appropriate (redundant) side information in following packets if the required delay to wait for this information is tolerable in the considered application.