Packet loss concealment in predictive audio coding

In this paper we present and evaluate several concealment strategies for packet losses in the context of a low delay predictive audio coder. Our goal is to minimize the audible impact of a packet loss. The problem is that the predictive coder is backward adaptive, hence depending on past values. There is a predictor reset, but to increase coding efficiency, the distance between two resets is several hundred packets. Hence, not only the lost packet itself cannot be reconstructed, but the transmitted data up to the next reset does not result in an exact reconstruction of the audio signal. Our approach is to try to use as much information as possible from the still available data until the next reset, and to reconstruct an audio signal such that distortions are least objectionable. To compare different approaches, we conducted a listening test. The result is that an adaptive reconstruction filter works best in this context.